Book D
Book D
Book D
Cover image: This shows arrows representing a vector field and a corresponding contour map. The
vector field shows the direction of change for a model of two competing populations of animals. You
will meet this model in Unit 12.
This publication forms part of an Open University module. Details of this and other
Open University modules can be obtained from the Student Registration and Enquiry Service, The
Open University, PO Box 197, Milton Keynes MK7 6BJ, United Kingdom (tel. +44 (0)845 300 6090;
email [email protected]).
Alternatively, you may visit the Open University website at www.open.ac.uk where you can learn
more about the wide range of modules and packs offered at all levels by The Open University.
To purchase a selection of Open University materials visit www.ouw.co.uk, or contact Open
University Worldwide, Walton Hall, Milton Keynes MK7 6AA, United Kingdom for a brochure
(tel. +44 (0)1908 858779; fax +44 (0)1908 858787; email [email protected]).
Contents
Unit 12 Systems of differential equations 1
Introduction 3
1 Visualising systems of differential equations 4
1.1 Direction fields revisited 4
1.2 Pictures of solutions 5
1.3 Modelling populations of predators and prey 8
1.4 More examples 11
2 Equilibrium points 13
2.1 Finding equilibrium points 13
2.2 Stability of equilibrium points 15
2.3 Behaviour close to equilibrium 16
4 Phase portraits 36
Learning outcomes 44
Solutions to exercises 45
Introduction 61
Introduction 121
1 Solving partial differential equations 121
1.1 Introducing partial differential equations 122
1.2 Initial conditions and boundary conditions 124
1.3 Separation of variables 129
Index 165
Unit 12
Introduction
Systems of linear first-order differential equations were considered in
Unit 6, and for these systems we were able to find an explicit solution.
Here we consider qualitative, graphical methods that are applicable to all
systems of first-order differential equations, but are of greatest value for
those that we cannot solve explicitly, namely non-linear systems.
Recall from Unit 6 that the system of differential equations
ẋ = −y, ẏ = 2x + y,
is called linear because the right-hand side of each equation does not
contain the variables raised to a power (such as x2 ) or as the argument of a
non-linear function (such as sin x), or products of the variables (such
as xy). The systems of differential equations
ẋ = x2 + y, ẏ = y + t, (1)
and
ẋ = xy, ẏ = x2 y, (2)
are both non-linear systems of equations.
There is one difference between systems (1) and (2) that is important for
this unit. In equations (1) the independent variable t appears explicitly on
the right-hand side, whereas in equations (2) only the two dependent
variables x and y occur. Systems such as (2), where t does not appear
explicitly, are said to be autonomous. In this unit we will consider only
autonomous equations of the general form
ẋ = u(x, y), ẏ = v(x, y).
The methods developed in this unit are widely applicable to a wide variety
of situations as differential equation models are common. For this reason
we will generally consider the behaviour of systems of differential equations
without any specific context, but we do develop one context to illustrate
ideas. The situation that we develop is a model of the behaviour of two
interacting populations of animals, usually with one variable x = x(t) We use the notations x or x(t),
representing the number of individuals of a predator species, and the other ẋ or ẋ(t), etc., interchangeably
variable y = y(t) representing the number of individuals of its prey. to suit the context.
3
Unit 12 Systems of differential equations
4
1 Visualising systems of differential equations
Exercise 1
Use the methods of Unit 1 to find the particular solution of equation (3)
that satisfies the initial condition x = x0 when t = 0.
5
Unit 12 Systems of differential equations
For the purposes of this discussion, suppose that the populations are
evolving independently (perhaps on separate islands) with no interactions.
The reason for this is that it makes the equations easy to solve, so the
solutions can be compared with the graphical method that we are about to
introduce. So consider the equations
ẋ = 0.2x, ẏ = 0.3y. (4)
(If the variables x and y represent populations, then we must have x ≥ 0
and y ≥ 0 in order to be physically reasonable. Here we are interested in
developing general methods, so we do not impose this restriction.)
Note that each of equations (4) is in the form of the differential
equation (3) considered previously, so they both represent exponential
growth.
Equations (4) form a system of linear differential equations, which you met
in Unit 6. Using vector notation, the pair of populations may be
represented by the vector x = (x y)T . The system of equations (4) then
becomes the vector equation
% - % -
ẋ 0.2x
ẋ = = . (5)
ẏ 0.3y
Vector fields are discussed in It is helpful now to introduce the notion of a vector field, which is similar
more detail in Unit 15. to a direction field. In a plane, a direction field associates a direction
f (x, y) with each point (x, y), whereas a vector field associates a vector
u(x, y) with each point (x, y). The vector field associated with
equation (5) is given by
% -
0.2x
u(x, y) = , (6)
0.3y
so equation (5) becomes ẋ = u(x, y), which will sometimes be written more
simply as ẋ = u(x). We could use this definition to calculate a direction at
any point. For example, at the point (1, 2), the vector field has the
direction u(1, 2) = (0.2 × 1 0.3 × 2)T = (0.2 0.6)T . Calculating vectors
at several points enables us to construct Figure 3, which shows a plot of
this vector field where an arrow in the direction of each vector u is placed
with its midpoint at the point where the vector was calculated.
A direction field f (x, y) represents the slope of a particular solution of the
differential equation dy/dx = f (x, y) at the point (x, y). Similarly, u(x, y)
is the vector ẋi + ẏj that is tangential to a particular solution of
ẋ = u(x, y) at the point (x, y), because the slope of the tangent is
dy dy/dt ẏ
= = .
dx dx/dt ẋ
This suggests a geometric way of finding a particular solution of
Figure 3 A representation of equation (5): choose a particular starting point (x0 , y0 ), then follow the
the vector field u(x, y)
directions of the tangent vectors. (An exception, which we discuss later,
occurs when ẋ = ẏ = 0 at (X, Y ), so u(X, Y ) = 0 and there is no tangent
vector to follow.)
6
1 Visualising systems of differential equations
The essential difference between a direction field and a vector field is that
the former consists of line segments, and the latter consists of directed line
segments (which we indicate by arrows) whose lengths indicate the
magnitude of u(x, y). However, since the magnitudes of u(x, y) may vary
considerably and so make the diagram difficult to interpret, we often use
arrows of a fixed length to show the direction of a vector field. This is
acceptable because, in many cases, it is the direction of u(x, y) that is our
primary concern, rather than its magnitude. Consider Figure 3, where the
arrows become longer as distance from the origin increases, so if the arrows
are scaled so that the longer arrows do not overlap, then it is hard to
discern the direction of the arrows near the origin. To overcome this
problem, we use the convention that all arrows will be scaled to the same
length. The arrows shown in Figure 4 represent the same vector field but
use this scaling convention.
Using the methods of Unit 6, we can find the general solution of
equations (4) as
x(t) = Ce0.2t , y(t) = De0.3t , (7)
where C and D are constants.
This general solution gives the position (x(t), y(t)) at time t. It is
instructive to plot this solution for a range of time t for various values of
the constants C and D. This will show the particular solutions for various
initial conditions among the family of general solutions. These particular Figure 4 The vector field
solutions are overlaid on the vector field plot in Figure 4. The arrows on u(x, y) together with paths
the solution curves indicate the directions in which the curves are from the family of solutions
traversed with increasing time.
A solution curve along which the coordinates x and y vary as t increases is
called a phase path (or orbit). The (x, y)-plane containing the solution
curves is called the phase plane, and a diagram, such as Figure 4,
showing the phase paths is called a phase portrait. In other words, a
phase portrait is a collection of phase paths that illustrates the behaviour
of the differential equations.
You may have noticed in Figure 4 that the paths radiate outwards from the
origin in all directions. For this reason, we refer to the origin as a source. A source can occur at a point
other than the origin.
We now look at the phase paths for a similar system, for which
% -
−0.2x
u(x, y) = .
−0.3y
Again by using the methods of Unit 6, the general solution can be found as
x = Ce−0.2t , y = De−0.3t ,
where C and D are constants.
This solution is in terms of exponential functions with negative exponents,
so it represents exponential decay. As this general solution is the same as
equations (7) except for the change in sign of the multiples of t, we can see
that the paths in the phase plane are the same as before, but with the
arrows reversed.
7
Unit 12 Systems of differential equations
Another way of looking at this is to say that at each point in the phase
plane, the vector field that represents these equations points in the
opposite direction to the previous vector field. This gives the diagram
shown in Figure 5. Now the paths radiate inwards towards the origin, and
for this reason we refer to the origin as a sink.
Exercise 2
Write down and solve the system of differential equations ẋ = u(x, y) given
by the vector field
% -
x
u(x, y) = .
Figure 5 The vector field −y
and solution paths of a sink
You should find that if you multiply together your solutions for x(t) and
y(t) found in Exercise 2, the result is a constant. So the phase paths have
the equation xy = A, for some constant A. These paths are a family of
rectangular hyperbolas (by choosing A to be non-zero) or lines along the
axes (by choosing A to be zero). Hence we can sketch the phase portrait
for the vector field examined in Exercise 2 to be as shown in Figure 6. You
can see that the vast majority of the paths do not radiate into or out of
the origin. On these paths, a point initially travels towards the origin, but
eventually travels away from it again. The only paths that actually radiate
inwards towards or outwards from the origin are those along the x- and
y-axes. In this case we call the origin a saddle point (the paths look like
the contours around a saddle point as defined in Unit 7).
Figure 6 Phase paths for This unit is concerned with sketching phase portraits. Before moving on to
the vector field in Exercise 2, give general methods for doing this, we introduce some examples to which
which shows a saddle point at
we can apply our methods.
the origin
8
1 Visualising systems of differential equations
Exercise 3
Sketch the graph of the proportionate growth rate ẋ/x of rabbits as a
function of the population y of foxes, and the graph of the proportionate
growth rate ẏ/y of foxes as a function of the population x of rabbits.
Interpret these graphs.
9
Unit 12 Systems of differential equations
Lotka–Volterra equations
The evolution of two interacting populations x and y can be modelled
by the Lotka–Volterra equations
* y) * x)
ẋ = kx 1 − , ẏ = −hy 1 − (x ≥ 0, y ≥ 0), (10)
Y X
where x is the population of the prey and y is the population of the
predators, and k, h, X and Y are positive constants.
Modelling success
This model was one of the first successful applications of
mathematical models to biological systems. It was independently
proposed in 1925 by the American biophysicist Alfred Lotka and in
1926 by the Italian mathematician Vito Volterra.
Exercise 4
(a) Write down the vector field u(x, y) that corresponds to the
Lotka–Volterra equations.
(b) Now suppose that the variables x and y represent thousands of
individuals (so that x = 1 represents one thousand rabbits, for
example) and further suppose that the constants in equations (10)
have the values k = 1, h = 21 , X = 3 and Y = 2 . Complete the
following table.
x y u(x, y)
0 0
0 2
2 0
2 2
3 1
3 2
3 3
4 2
(c) Draw the vectors that you obtained in part (b) as a vector field in the
phase plane.
10
1 Visualising systems of differential equations
formulas for x(t) and y(t) are available. We will use graphical methods to
describe the solutions. The vector field plot that you drew in Exercise 4
(partly reproduced as Figure 7(a)) is a starting point. From this plot it
may seem possible that the phase path forms a closed loop about the point
(3, 2), but this is far from certain. Figure 7(b) shows a sketch of a possible
phase path.
Figure 7 (a) Part of the vector field drawn in Exercise 4. (b) A possible
path for the changes in rabbit and fox populations that follows the vector
field.
11
Unit 12 Systems of differential equations
Exercise 5
Convert each of the following differential equations into a system of
first-order equations.
d2 x dx d3 u du
(a) ẍ + sin x = 0 (b) 2
−2 + x2 = 0 (c) 3
= 6u
dt dt dx dx
Using this method you can see that any situation that leads to a
differential equation model (such as any dynamics problem) can be
converted into a form to which we can apply the graphical methods
described in this unit. These methods are widely applicable.
Note that not every system of first-order differential equations can be
converted into a single higher-order differential equation by reversing the
above strategy. For example, the Lotka–Volterra equations cannot be
written as a single second-order differential equation in this way.
This strategy of converting a higher-order differential equation to a system
of first-order differential equations is also useful in the numerical solution
of differential equations. Many numerical methods (e.g. Euler’s method,
which you met in Unit 1) apply only to first-order differential equations.
These methods can be extended in a straightforward manner to apply to
systems of first-order equations (by replacing scalar variables with vector
variables), and this is the main method for solving higher-order differential
equations on a computer. The numerical solutions of the weather
prediction model employed to produce daily weather forecasts use exactly
this strategy.
Now we have plenty of examples of systems of first-order differential
equations and a graphical picture (the phase plane) to represent them. We
go on to analyse the features of the graphical picture. First, we look at
steady-state solutions of the equations (such as the point (3, 2) in
Exercise 4) where two populations can coexist in equilibrium.
12
2 Equilibrium points
2 Equilibrium points
The previous section introduced phase portraits as a method of visualising
systems of non-linear differential equations. Now we begin to investigate
features of phase plane portraits, and we start with the features called
equilibrium points.
These single points, such as the origin in Figure 8, are important features
of phase portraits, so we make the following definition.
13
Unit 12 Systems of differential equations
Example 1
Find the equilibrium points for the Lotka–Volterra equations (10) for the
rabbit and fox populations. (Remember that h, k, X and Y are positive
constants.)
Solution
Using Procedure 1, we need to solve the equation u(x, y) = 0, which
becomes
* y ) % -
kx 1 −
Y = 0 .
* x ) 0
−hy 1 −
X
This gives the simultaneous equations
* y)
kx 1 − = 0, (12)
Y
* x)
−hy 1 − = 0. (13)
X
It is important to be methodical when solving simultaneous non-linear
equations, in order to avoid missing solutions. Here we will first solve
equation (12) and then substitute each solution into equation (13) to find
all solutions. Equation (12) is already factorised and the solutions arise
when either term is zero, so the solutions are x = 0 or y = Y .
Substituting x = 0 into equation (13) gives −hy = 0, so y = 0 and hence
(0, 0) is an equilibrium point.
Substituting y = Y into equation (13) gives −hY (1 − x/X) = 0, so x = X
and hence (X, Y ) is an equilibrium point. As both X and Y are positive,
this equilibrium point is always in the first quadrant (thus is always in the
quadrant x > 0, y > 0 that is physically relevant for population models).
Thus there are two possible equilibrium points for the pair of populations.
The first has both the rabbit and fox populations zero, that is, the
equilibrium point is at (0, 0); there are no births or deaths – nothing
happens. However, the other equilibrium point occurs when there are
14
2 Equilibrium points
X rabbits and Y foxes, that is, the equilibrium point is at (X, Y ); the This explains our choice of
births and deaths exactly cancel out and both populations remain constants X and Y in
constant. Subsection 1.3.
Exercise 6
Suppose that two variables x and y evolve according to the system of
differential equations
ẋ = x(20 − y), ẏ = y(10 − y)(10 − x).
Find the equilibrium points of the system.
Exercise 7
Find the equilibrium points of the system of differential equations
ẋ = 2x2 y + 7xy 2 + 2y + 1,
ẏ = xy − x.
15
Unit 12 Systems of differential equations
On the other hand, the origin in the phase portrait of a source shown in
Figure 4 is an unstable equilibrium point. Any perturbation from the
origin will result in the point travelling further and further away from the
origin with time. Similarly, the origin in the phase portrait of a saddle
shown in Figure 6 is an unstable equilibrium point. Apart from increases
or decreases in y with x unchanged, any perturbation will result in a point
that travels further and further away from the origin with time.
Exercise 8
Consider the paths shown in Figure 10.
16
2 Equilibrium points
17
Unit 12 Systems of differential equations
Now, for small perturbations p and q, we can use the linear Taylor
polynomial for functions of two variables to approximate each of u(x, y)
and v(x, y) near the equilibrium point (X, Y ). Here we use the more
compact notation ux for ∂u/∂x etc.:
u(X + p, Y + q) & u(X, Y ) + p ux (X, Y ) + q uy (X, Y )
= p ux (X, Y ) + q uy (X, Y ),
since u(X, Y ) = 0, and
v(X + p, Y + q) & v(X, Y ) + p vx (X, Y ) + q vy (X, Y )
= p vx (X, Y ) + q vy (X, Y ),
since v(X, Y ) = 0.
The above two equations appear rather unwieldy, but are much more
succinctly represented in matrix form:
% - % -% -
u(x, y) ux (X, Y ) uy (X, Y ) p
= .
v(x, y) vx (X, Y ) vy (X, Y ) q
Since x(t) = X + p(t) and y(t) = Y + q(t), we also have
ẋ = ṗ, ẏ = q̇.
Putting the pieces together, substituting in ẋ = u(x, y) gives a system of
linear differential equations for the perturbations p and q:
% - % -% -
ṗ ux (X, Y ) uy (X, Y ) p
= . (18)
q̇ vx (X, Y ) vy (X, Y ) q
Some examples will help to make this clear.
Example 2
Suppose that two variables x and y evolve according to the system of
differential equations
The equilibrium points for these ẋ = x(20 − y), ẏ = y(10 − y)(10 − x).
equations were found in
Exercise 6. Find the linear approximation to these equations near the equilibrium
point (0, 10).
Solution
Here we have
u(x, y) = x(20 − y), v(x, y) = y(10 − y)(10 − x).
So the partial derivatives are
ux (x, y) = 20 − y, uy (x, y) = −x,
vx (x, y) = −y(10 − y), vy (x, y) = (10 − y)(10 − x) − y(10 − x).
Evaluating these at the given point (0, 10) yields
ux (x, y) = 10, uy (x, y) = 0,
vx (x, y) = 0, vy (x, y) = −100.
18
2 Equilibrium points
So the linear system that approximates the given system near the point
(0, 10) is
% - % -% -
ṗ 10 0 p
= .
q̇ 0 −100 q
Example 3
Transform the Lotka–Volterra equations (10) into a system of linear
differential equations for the perturbations p and q from the equilibrium
point (X, Y ).
Solution
Here we have
* y) * x)
u(x, y) = kx 1 − , v(x, y) = −hy 1 − .
Y X
First, we compute the partial derivatives, obtaining
* y) kx
ux (x, y) = k 1 − , uy (x, y) = − ,
Y Y
hy * x)
vx (x, y) = , vy (x, y) = −h 1 − .
X X
Evaluating these at the point (X, Y ) gives
kX
ux (X, Y ) = 0, uy (X, Y ) = − ,
Y
hY
vx (X, Y ) = , vy (X, Y ) = 0.
X
Thus the required system of linear differential equations is
% - % -% -
ṗ 0 −kX/Y p
= . (19)
q̇ hY /X 0 q
19
Unit 12 Systems of differential equations
Exercise 9
Write down the linear approximations to the Lotka–Volterra
equations (10) near the equilibrium point (0, 0).
Exercise 10
This system was also considered Consider the equations
in Exercise 6 and Example 2.
ẋ = x(20 − y), ẏ = y(10 − y)(10 − x).
Find the linear approximations to these equations near the equilibrium
point (10, 20).
Exercise 11
Find the equilibrium point of the system of differential equations
ẋ = 3x + 2y − 8, ẏ = x + 4y − 6.
Find a system of linear differential equations satisfied by small
perturbations p and q from the equilibrium point.
20
3 Classifying equilibrium points
Exercise 12
Suppose that the pair of populations x and y can be modelled by the
system of differential equations
ẋ = 0.5x − 0.000 05x2 ,
ẏ = −0.1y + 0.0004xy − 0.01y 2
(x ≥ 0, y ≥ 0).
(a) Find the three equilibrium points of the system.
(b) Find the Jacobian matrix of the system.
(c) For each of the three equilibrium points, find the linear differential
equations that give the approximate behaviour of the system near the
equilibrium point.
22
3 Classifying equilibrium points
Exercise 14
Consider the linear system of differential equations
% - % -% -
ṗ 0 −1 p
= .
q̇ 2 −3 q
(a) Find the eigenvalues of the matrix of coefficients.
(b) Classify the equilibrium point p = 0, q = 0 of the system.
So far in this section we have considered the case where the matrix of
coefficients has two distinct positive eigenvalues and the case where the
matrix has two distinct negative eigenvalues. We now consider the case
where the matrix has one positive eigenvalue and one negative eigenvalue.
For example, consider the matrix
% -
1 4
J= ,
1 −2
which has eigenvalues 2 and −3, and corresponding eigenvectors (4 1)T
and (1 −1)T . The general solution of the linear system of differential
equations ṗ = Jp is
% - % - % -
p 4 2t 1 −3t
=C e +D e , (24)
q 1 −1
where C and D are constants. When D = 0 (and C '= 0), we have
p(t) = 4Ce2t and q(t) = Ce2t , and the point (p(t), q(t)) moves away from
the origin along the straight-line path q = 41 p as t increases; this line is an
eigenline. On the other hand, when C = 0 (and D '= 0), the solution is
p(t) = De−3t , q(t) = −De−3t , so the point (p(t), q(t)) approaches the origin
along the straight-line path q = −p as t increases.
Hence we have seen that there are two straight-line paths. On the line
q = 41 p (which corresponds to the eigenvector (4 1)T ), the point moves
away from the origin as t increases. However, on the line q = −p (which
corresponds to the eigenvector (1 −1)T ), the point moves towards the Figure 13 Paths to and
origin as t increases. These paths are shown in Figure 13. from an equilibrium point
23
Unit 12 Systems of differential equations
Now we consider the behaviour of a general point (p(t), q(t)), where p(t)
and q(t) are given by equation (24), and neither C nor D is zero. For large
positive values of t, the terms involving e2t dominate, so p(t) & 4Ce2t and
q(t) & Ce2t . So for large positive values of t, the general path approaches
the line q = 14 p. On the other hand, for large negative values of t, the
terms involving e−3t dominate, so p(t) & De−3t and q(t) & −De−3t . So for
large negative values of t, the general path approaches the line q = −p.
Using this information we can add to Figure 13 to obtain Figure 14.
We can see that the equilibrium point is a saddle. The type of behaviour
shown in Figure 14 occurs when the matrix of coefficients has one positive
Figure 14 Paths near a eigenvalue and one negative eigenvalue. Again, the straight-line paths are
saddle point; the dot marks in the directions of the eigenvectors of the matrix.
the equilibrium point
Exercise 15
Consider the linear system of differential equations
% - % -% -
ṗ 1 2 p
= .
q̇ 2 −2 q
(a) Find the eigenvalues and corresponding eigenvectors of the matrix of
coefficients.
(b) Classify the equilibrium point p = 0, q = 0.
(c) Sketch the phase paths of the solutions of the differential equations in
the neighbourhood of (0, 0).
Example 4
Consider the linear system of differential equations
% - % -% -
ṗ 0 −1 p
= .
q̇ 4 0 q
(a) Find the eigenvalues and corresponding eigenvectors of the matrix of
coefficients.
(b) Hence write down the general solution of the system of differential
equations.
(c) Show that the phase paths for these differential equations are the
ellipses
p2 + 41 q 2 = K,
where K is a positive constant.
24
3 Classifying equilibrium points
Solution
(a) The matrix of coefficients has characteristic equation
' '
'−λ −1 '
' '
' 4 −λ' = 0,
In Example 4, we saw that all the phase paths are ellipses. This type of
behaviour corresponds to any linear system of differential equations where Figure 15 An elliptical path
the eigenvalues of the matrix of coefficients are purely imaginary. An in the phase plane near a
equilibrium point that has this behaviour in its neighbourhood is called a centre
centre.
25
Unit 12 Systems of differential equations
Exercise 16
Consider the linear system of differential equations
% - % -% -
ṗ 2 −1 p
= .
q̇ 5 −2 q
(a) Find the eigenvalues of the matrix of coefficients.
(b) Classify the equilibrium point p = 0, q = 0.
In general, when the eigenvalues of a matrix are complex, they are not
purely imaginary but also contain a real part. This has a significant effect
on the solution of the corresponding system, as you will see in the
following example.
Example 5
Find the general solution of the system of equations ṗ = Jp, where
% -
−2 −3
J= .
3 −2
Sketch some paths corresponding to the solutions of the system.
Solution
The characteristic equation of the matrix of coefficients is (2 + λ)2 + 9 = 0,
so the eigenvalues are −2 + 3i and −2 − 3i. Corresponding eigenvectors
are (1 −i)T and (1 i)T , respectively, so the general solution is given by
% - % - % -
p −2t cos 3t −2t sin 3t
= Ce + De ,
q sin 3t − cos 3t
where C and D are constants.
If we neglect, for the time being, the e−2t terms, the solution is
p = C cos 3t + D sin 3t,
q = C sin 3t − D cos 3t,
from which it follows that
p2 + q 2 = C 2 + D 2 .
So in the absence of the e−2t terms, the paths would be circles with centre
at the origin. The effect of the e−2t terms on these paths is to reduce the
radius of the circles gradually. In other words, the paths spiral in towards
the origin as t increases, as shown in Figure 16.
Figure 16 Paths near a In Example 5 (and Figure 16) the paths spiral in towards the origin, so the
spiral sink ; the dot marks the origin is a sink (called a spiral sink) and therefore is a stable equilibrium
equilibrium point point.
26
3 Classifying equilibrium points
Exercise 17
Consider the linear system of differential equations Figure 17 Paths near a
% - % -% -
ṗ 1 1 p spiral source; the dot marks
= . the equilibrium point
q̇ −1 1 q
(a) Find the eigenvalues of the matrix of coefficients.
(b) Classify the equilibrium point p = 0, q = 0.
Exercise 18
Consider the linear system of differential equations
% - % -% -
ṗ 2 0 p
= .
q̇ 0 2 q
(a) Find the eigenvalues and eigenvectors of the coefficient matrix.
(b) Find the general solution of the system of differential equations.
(c) By eliminating t, find the equations of the paths, and describe them.
(d) Is the equilibrium point p = 0, q = 0 stable or unstable?
27
Unit 12 Systems of differential equations
In Exercise 18, we have seen that when the matrix of coefficients has two
real identical positive eigenvalues and two linearly independent
eigenvectors, all the paths are straight lines radiating away from the origin,
as shown in Figure 18. The equilibrium point at p = 0, q = 0 is called a
star source. If there are two identical negative eigenvalues (but still two
linearly independent eigenvectors), then the arrows on the paths in
Figure 18 are reversed, and the equilibrium point at p = 0, q = 0 is called a
star sink.
We turn finally to the case where there are two identical eigenvalues but
only one independent eigenvector.
Figure 18 Paths near a star
source; the dot marks the
equilibrium point Exercise 19
(a) Find the eigenvalues and corresponding eigenvectors of the matrix
% -
2 0
J= .
1 2
(b) Find the general solution of the system of differential equations
ṗ = Jp.
28
3 Classifying equilibrium points
29
Unit 12 Systems of differential equations
Exercise 20
In Example 3 we saw that the Lotka–Volterra equations can be
approximated by the system of linear differential equations
% - % -% -
ṗ 0 −kX/Y p
=
q̇ hY /X 0 q
in the neighbourhood of the equilibrium point (X, Y ). Find the eigenvalues
of the matrix of coefficients, and hence classify the equilibrium point p = 0,
q = 0.
Exercise 21
In Exercise 9 we saw that the Lotka–Volterra equations can be
approximated by the system of linear differential equations
% - % -% -
ṗ k 0 p
=
q̇ 0 −h q
in the neighbourhood of the equilibrium point (0, 0). Find the eigenvalues
of the matrix of coefficients, and hence classify the equilibrium point p = 0,
q = 0.
30
3 Classifying equilibrium points
31
Unit 12 Systems of differential equations
Example 6
Consider the non-linear system of differential equations
ẋ = −4y + 2xy − 8, ẏ = 4y 2 − x2 .
(a) Find the equilibrium points of the system.
(b) Compute the Jacobian matrix of the system.
(c) In the neighbourhood of each equilibrium point:
• linearise the system of differential equations
• classify the equilibrium point of the linearised system.
(d) What can you say about the classification of the equilibrium points of
the original (non-linear) system of differential equations?
Solution
(a) The equilibrium points are given by
−4y + 2xy − 8 = 0,
4y 2 − x2 = 0.
The second equation gives
x = ±2y.
When x = 2y, substitution into the first equation gives
−4y + 4y 2 − 8 = 0,
or y 2 − y − 2 = 0, which factorises to give
(y − 2)(y + 1) = 0.
32
3 Classifying equilibrium points
Hence
y = 2 or y = −1.
When y = 2, x = 2y = 4. When y = −1, x = 2y = −2. So we have
found two equilibrium points, namely (4, 2) and (−2, −1).
When x = −2y, substitution into the first equation gives
−4y − 4y 2 − 8 = 0,
or y 2 + y + 2 = 0. This quadratic equation has no real solutions, so
there are no more equilibrium points.
(b) Differentiating the right-hand sides of the given differential equations
gives the Jacobian matrix as
% -
2y 2x − 4
J= .
−2x 8y
(c) At the equilibrium point (4, 2), the Jacobian matrix is
% -
4 4
,
−8 16
so the linearised system is
% - % -% -
ṗ 4 4 p
= .
q̇ −8 16 q
The characteristic equation of the matrix of coefficients is
(4 − λ)(16 − λ) + 32 = 0,
or λ2 − 20λ + 96 = 0, which factorises to give
(λ − 8)(λ − 12) = 0,
so the eigenvalues are
λ = 8 and λ = 12.
The two eigenvalues are positive and distinct, so the equilibrium point
p = 0, q = 0 is a source.
At the equilibrium point (−2, −1), the Jacobian matrix is
% -
−2 −8
,
4 −8
so the linearised system is
% - % -% -
ṗ −2 −8 p
= .
q̇ 4 −8 q
The characteristic equation of the matrix of coefficients is
(−2 − λ)(−8 − λ) + 32 = 0,
which simplifies to
λ2 + 10λ + 48 = 0.
33
Unit 12 Systems of differential equations
Exercise 22
A certain system of differential equations has an equilibrium point where
the Jacobian matrix evaluates to
% -
2 −3
J= .
3 2
(a) Find the eigenvalues and eigenvectors of this matrix, and hence
classify the equilibrium point.
(b) If this Jacobian matrix is a linear approximation to a non-linear
system ẋ = u(x, y), what can you say about the equilibrium point of
the non-linear system?
Exercise 23
Consider the non-linear system of differential equations
ẋ = (1 + x − 2y)x, ẏ = (x − 1)y.
(a) Find the equilibrium points of the system.
(b) Find the Jacobian matrix of the system.
(c) In the neighbourhood of each equilibrium point:
• find the linear system of differential equations that gives the
approximate behaviour of the system near the equilibrium point
• find the eigenvalues of the matrix of coefficients
• use the eigenvalues to classify the equilibrium point of the
linearised system.
(d) What can you say about the classification of the equilibrium points of
the original non-linear system of differential equations?
34
3 Classifying equilibrium points
4 Phase portraits
Exercise 24
Consider the system of differential equations defined in Exercise 4, namely
* y) 1 * x)
ẋ = x 1 − , ẏ = − y 1 − .
2 2 3
As we are interested in general features of these equations, consider both
positive and negative values of x and y rather than restricting attention to
those values that are physically reasonable for a population model
(i.e. x ≥ 0 and y ≥ 0).
(a) For what values of x and y is ẋ zero? In what regions of the plane is ẋ
positive? Sketch these regions on the phase plane.
(b) For what values of x and y is ẏ zero? In which regions is ẏ positive?
Sketch these regions on the phase plane.
The regions of the plane that you sketched in Exercise 24 are significant.
The regions where ẋ is positive are the regions where the vector field
arrows are pointing generally to the right. In these regions the phase paths
will curve to the right. Similarly, the regions where ẏ is positive are the
regions where the phase paths curve upwards. This information is really
useful when sketching phase portraits.
The boundaries of the regions are the key concept, as these boundaries are
the only places where curves can change between curving
leftwards/rightwards and upwards/downwards, so these are given a name.
36
4 Phase portraits
Now we will use nullclines to plot the phase portrait of the predator–prey
equations. The process that we go through will later be formalised in a
procedure, and we will mark the steps of the procedure in the margin as
we go along.
Example 7
Sketch the phase portrait of the system of differential equations
* y) 1 * x)
ẋ = x 1 − , ẏ = − y 1 − .
2 2 3
Solution
The first step in sketching a phase portrait is to find and classify the " Equilibrium points !
equilibrium points. In Example 1 we found that these equations have two
equilibrium points, in this case (with X = 3 and Y = 2) (0, 0) and (3, 2).
The given equations are the Lotka–Volterra equations for specific
parameters; in Exercises 20 and 21 we classified the equilibrium points and
found that (0, 0) is a saddle point and (3, 2) is a centre. These are marked
in green in Figure 23 (see below).
The next step is to find the nullclines. In Exercise 24 we found that the " Nullclines !
nullclines for ẋ are x = 0 and y = 2. These are marked by red lines in
Figure 23. Also in Exercise 24 we found that the nullclines for ẏ are y = 0
and x = 3. These are marked by blue lines in Figure 23.
Phase paths cross nullclines for ẋ vertically (since ẋ = 0). We need to " Nullcline crossings !
determine whether each crossing is upwards or downwards – this is
determined by the sign of ẏ on the nullcline. This sign can change only at
the equilibrium points, as the right-hand sides of the system of differential
equations are continuous. So all we need to do is determine the sign of ẏ
on either side of each equilibrium point. To do this, we compute the
following values.
(x, y) (0, 1) (0, −1) (2, 2) (4, 2)
ẏ −1/2 1/2 −1/3 1/3
In Figure 23, the positive signs are represented by upwards arrows, and the
negative signs are represented by downward arrows.
Similarly, the phase paths cross the nullclines for ẏ horizontally. We can
evaluate ẋ at points on either side of equilibrium points to find the
direction of crossing as follows.
(x, y) (1, 0) (−1, 0) (3, 1) (3, 3)
ẋ 1 −1 3/2 −3/2
37
Unit 12 Systems of differential equations
Again, we represent the signs of these values by arrows in Figure 23, where
positive signs are represented by rightwards arrows and negative signs are
represented by leftwards arrows.
" Nullcline regions ! The nullclines divide the phase plane into regions. In each region the
vector field arrows will point in the same general direction (up/down and
left/right) as ẋ and ẏ can change sign only on the nullclines. In Exercise 24
the regions shaded red are the regions where ẋ > 0, and the regions shaded
blue are the regions where ẏ > 0. So the overlap regions are the regions
where ẋ > 0 and ẏ > 0, that is, the arrows will point rightwards and
upwards. Similarly, the regions that are not shaded by either colour are
regions where ẋ < 0 and ẏ < 0, that is, regions where the arrows point
leftwards and downwards. These signs are represented by arrows pointing
We often abbreviate these north-east, north-west, south-east and south-west in Figure 23. Note that
directions to NE, NW, SE the arrows in this figure are purely representative of the general direction of
and SW. the arrows in each region; they are not the computed arrows at this point.
" Complete paths ! All that remains is to use this information to sketch typical paths in the
phase plane. The aim here is not to fill the phase plane with paths, but to
add sufficient paths to show the important features. This will usually
involve showing the paths that start at equilibrium points (i.e. start
infinitesimally close to but away from equilibrium points) and exploring
each of the regions defined by the nullclines. The phase portrait for these
equations is shown in Figure 24.
38
4 Phase portraits
The steps used in Example 7 form the basis of a general procedure for
sketching phase portraits.
39
Unit 12 Systems of differential equations
Example 8
Consider the system of differential equations that we looked at in
Example 6, namely
ẋ = −4y + 2xy − 8, ẏ = 4y 2 − x2 ,
where we found that these equations have a source at (4, 2) and a spiral
" Equilibrium points ! sink at (−2, −1).
Sketch the phase portrait.
Solution
" Nullclines ! The nullclines for ẋ are given by
−4y + 2xy − 8 = 0.
Rearranging to make y the subject of this equation gives
4
y= .
x−2
So these nullclines are the two branches of a rectangular hyperbola
obtained by taking the graph of y = 1/x, scaling by a factor 4, and
translating 2 units to the right.
Now consider the nullclines for ẏ, which are given by
4y 2 − x2 = 0.
This equation has solutions x = 2y and x = −2y. So the nullclines in this
case are a pair of straight lines.
A sketch of the nullclines is shown in Figure 25(a). One interesting feature
to note from the diagram is that the equilibrium points (marked by dots)
occur at the intersections of the red and blue lines. This will always
happen, as the red lines correspond to ẋ = 0 and the blue lines correspond
to ẏ = 0, so the intersection points are when both ẋ = 0 and ẏ = 0, that is,
the equilibrium points.
" Nullcline crossings ! Now we determine the directions of the nullcline crossings. We start with
the nullclines for ẋ, which are the two branches of the rectangular
hyperbola. Both equilibrium points lie on these nullclines, so we need to
evaluate ẏ at four points (one on each side of each equilibrium point). We
choose the points to simplify the arithmetic.
(x, y) (−6, −1/2) (0, −2) (3, 4) (6, 1)
ẏ −35 16 55 −32
The signs of these values are represented by red upwards and downwards
pointing arrows in Figure 25(b).
Now consider the nullclines for ẏ. The nullcline x = 2y also has two
equilibrium points lying on it, but this time we can reduce the effort by
considering only three points as the curve is continuous. So we calculate
the following.
40
4 Phase portraits
The other nullcline for ẏ, namely the line x = −2y, has no equilibrium
points on it. So the direction of crossing must be the same along the whole
length of this line. So we need to evaluate ẋ at only one point to determine
this sign. In fact, we do not even have to do one evaluation, as this
nullcline crosses the other nullcline for ẏ, and we know that the sign at the
point of crossing is negative. So the sign of ẏ must be negative along the
whole length of this nullcline. This is marked in Figure 25(b).
We also know that the point (4, 2) is a source and the point (−2, −1) is a
spiral sink. So the solution paths will flow outwards from the point (4, 2)
and spiral inwards to the point (−2, −1). This information has also been
added to Figure 25(b) as small sketches of the phase paths in the
neighbourhoods of the equilibrium points. It helps to delay adding these
sketches in the neighbourhoods of the equilibrium points until the nullcline
crossings are known, as this helps to determine the sketch; for example, it
indicates whether a spiral sink is a clockwise or anticlockwise spiral.
We now have a lot of information about the directions of the solution paths " Nullcline regions !
on the nullclines, and we can extend this to the regions of the plane with
the nullclines as boundaries. Consider the region between the two branches
of the red hyperbola. As the red hyperbola corresponds to the equation
ẋ = 0, we know that ẋ cannot change sign in this region, so all solution
paths must progress leftwards. Similarly, we know that all solution paths
in the top-right and bottom-left regions must progress to the right. Also,
the plane is divided into four regions by the pair of blue nullclines. The
solution paths must be generally pointing upwards in the north and south
regions, and generally pointing downwards in the west and east regions.
These are represented by arrows in the regions in Figure 26.
41
Unit 12 Systems of differential equations
" Complete paths ! Now that we have gathered all the information about the system of
differential equations, we can sketch the phase portrait, as shown in
Figure 27. You should not try to be too precise when drawing a sketch
phase portrait. The aim is to convey the main features of the paths rather
than precise details.
Exercise 25
Exercise 23 considered the system of differential equations
ẋ = (1 + x − 2y)x, ẏ = (x − 1)y,
and found that this system has three equilibrium points, namely a saddle
point at (0, 0), a sink at (−1, 0) and a spiral source at (1, 1).
" Equilibrium points ! (a) Mark the equilibrium points on a sketch, and draw short paths in the
neighbourhood of each equilibrium point.
42
4 Phase portraits
(b) Find the nullclines, and add these to your sketch. " Nullclines !
(c) Find the signs of the paths crossing the nullclines, and mark these on " Nullcline crossings !
your sketch.
(d) Add NW, NE, SW or SE arrows to each region of the phase plane " Nullcline regions !
separated by the nullclines on your sketch.
(e) Sketch the phase portrait of this system of equations. " Complete sketch !
This section concludes by looking ahead to phenomena that you may meet
in your studies after completing this module. The first of these phenomena
is when a solution tends to a repeating pattern rather than tending to an
equilibrium point.
Recall that periodic solutions are represented by closed curves in the phase
plane. Here we name a particular type of periodic solution.
43
Unit 12 Systems of differential equations
Learning outcomes
After studying this unit, you should be able to:
• use a vector field to describe a pair of first-order non-linear differential
equations, and use paths in the phase plane to represent the solutions
• understand how systems of differential equations arise from
mathematical modelling, and in particular the modelling of
populations of predators and prey
• convert a higher-order differential equation to a system of first-order
differential equations
• find the equilibrium points of a system of non-linear differential
equations
• find linear equations that approximate the behaviour of a system of
non-linear differential equations near an equilibrium point, by
calculating the Jacobian matrix
• determine whether an equilibrium point is stable or unstable
• use the eigenvalues and eigenvectors of the Jacobian matrix at an
equilibrium point to classify an equilibrium point as a source, a sink, a
star source, a star sink, an improper source, an improper sink, a spiral
source, a spiral sink, a saddle, or a centre
• use nullclines to sketch the phase portrait of a system of differential
equations.
44
Solutions to exercises
Solutions to exercises
Solution to Exercise 1
The differential equation ẋ = kx can be solved using the method of
separation of variables to get, as x > 0,
& &
dx
= k dt.
x
Evaluating the integrals (and noting that x is positive) gives
ln x = kt + A,
where A is a constant.
Using the given initial condition determines the value of the constant A:
ln x0 = k × 0 + A.
Substituting for A and simplifying gives the required particular solution
ln x = kt + ln x0 .
We can write this as
ln x − ln x0 = kt or ln(x/x0 ) = kt,
so
x = x0 ekt .
Solution to Exercise 2
The system of differential equations is
ẋ = x, ẏ = −y.
These equations can be solved applying separation of variables to get the
two equations
& & & &
1 1
dx = 1 dt, dy = −1 dt.
x y
Performing the integrations and rearranging gives the general solution as
x(t) = Cet , y(t) = De−t ,
where C and D are constants.
45
Unit 12 Systems of differential equations
Solution to Exercise 3
Rearranging equation (8), we obtain
ẋ k
= k − y,
x Y
which is the equation of a straight line, as shown in the margin.
The proportionate growth rate ẋ/x of rabbits decreases as the
population y of foxes increases, becoming zero when y = Y . The
population x of rabbits will increase if the population y of foxes is less
than Y , but will decrease if y is greater than Y .
Similarly rearranging equation (9), we have
ẏ h
= −h + x,
y X
so the graph of ẏ/y as a function of x is as shown in the margin.
The proportionate growth rate ẏ/y of foxes increases linearly as the
population x of rabbits increases. The population y of foxes will decrease
if the population x of rabbits is less than X, but will increase if x is
greater than X.
Solution to Exercise 4
(a) The Lotka–Volterra equations can be written as
ẋ = u(x, y),
where ẋ = (ẋ ẏ)T and the vector field u(x, y) is given by
* y )
kx 1 −
Y .
u(x, y) = * x )
−hy 1 −
X
(b) The completed table is shown below.
x y u(x, y)
0 0 0
0 2 (0 −1)T
2 0 (2 0)T
2 2 (0 −1/3)T
3 1 (3/2 0)T
3 2 0
3 3 (−3/2 0)T
4 2 (0 1/3)T
46
Solutions to exercises
Solution to Exercise 5
(a) Let y = ẋ, so ẏ = ẍ. Then the given equation becomes ẏ + sin x = 0.
This leads to the system of first-order equations
ẋ = y, ẏ = − sin x.
(b) As before, let y = ẋ, so ẏ = ẍ. Then the given equation becomes
ẏ − 2y + x2 = 0, and the equivalent first-order system is
ẋ = y, ẏ = 2y − x2 .
(c) Here we have different variables and a third-order equation, but the
basic idea is still the same – that is, we introduce new variables for the
derivatives of the independent variable. So let
du d2 u dv
v= and w = 2 = .
dx dx dx
In terms of these variables the given equation becomes
dw
= 6uv.
dx
So the equivalent system of first-order equations is
du dv dw
= v, = w, = 6uv.
dx dx dx
Solution to Exercise 6
Procedure 1 leads to the pair of simultaneous equations
x(20 − y) = 0,
y(10 − y)(10 − x) = 0.
From the first equation, either x = 0 or y = 20.
Substituting x = 0 into the second equation gives y(10 − y) × 10 = 0,
which gives y = 0 or y = 10. So (0, 0) and (0, 10) are equilibrium points.
Substituting y = 20 into the second equation gives
20 × (−10) × (10 − x) = 0, which has solution x = 10. So (10, 20) is an
equilibrium point.
Hence the complete list of equilibrium points is (0, 0), (0, 10) and (10, 20).
47
Unit 12 Systems of differential equations
Solution to Exercise 7
Using Procedure 1, we must solve the pair of simultaneous equations
2x2 y + 7xy 2 + 2y + 1 = 0,
xy − x = 0.
In order to reduce the effort needed, we start by considering solutions of
the second equation as it is much simpler. The second equation factorises
as x(y − 1) = 0, which gives x = 0 or y = 1.
Substituting x = 0 into the first equation gives 2y + 1 = 0, which has
solution y = − 21 . So (0, − 12 ) is an equilibrium point.
Substituting y = 1 into the first equation gives 2x2 + 7x + 3 = 0, which
factorises as (2x + 1)(x + 3) = 0. So x = − 12 or x = −3, which gives the
equilibrium points (− 21 , 1) and (−3, 1).
So the complete list of equilibrium points is (0, − 12 ), (− 12 , 1) and (−3, 1).
Solution to Exercise 8
The paths move towards the equilibrium point A, so it is stable.
The paths move away from the equilibrium point B, so it is unstable.
The paths move around the equilibrium point C thus stay in the
neighbourhood of the point, so it is stable.
Solution to Exercise 9
We evaluate the various partial derivatives found in Example 3. At the
equilibrium point (0, 0), we obtain
ux (0, 0) = k, uy (0, 0) = 0,
vx (0, 0) = 0, vy (0, 0) = −h.
Thus the required linear approximation is
% - % -% -
ṗ k 0 p
= ,
q̇ 0 −h q
giving the pair of equations
ṗ = kp, q̇ = −hq.
Solution to Exercise 10
Here we have
u(x, y) = x(20 − y), v(x, y) = y(10 − y)(10 − x),
giving partial derivatives
ux (x, y) = 20 − y, uy (x, y) = −x,
vx (x, y) = −y(10 − y), vy (x, y) = (10 − y)(10 − x) − y(10 − x).
So the Jacobian matrix of the vector field u(x, y) is
% -
20 − y −x
J(x, y) = .
−y(10 − y) (10 − y)(10 − x) − y(10 − x)
48
Solutions to exercises
Solution to Exercise 11
Solving the equations 3x + 2y − 8 = 0 and x + 4y − 6 = 0, we obtain the
equilibrium point (2, 1).
Putting x = 2 + p and y = 1 + q, we obtain the matrix equation
% - % -% -
ṗ 3 2 p
= .
q̇ 1 4 q
(Note that this linear approximation near (2, 1) is exact, since the original
system is linear.)
Solution to Exercise 12
(a) To find the equilibrium points, we solve the simultaneous equations
0.5x − 0.000 05x2 = 0,
−0.1y + 0.0004xy − 0.01y 2 = 0.
Factorising these equations gives
0.5x(1 − 0.0001x) = 0,
−0.1y(1 − 0.004x + 0.1y) = 0.
The first equation gives
x=0 or x = 10 000.
If x = 0, the second equation is
−0.1y(1 + 0.1y) = 0,
which gives y = 0 or y = −10. As y ≥ 0, only the first solution is
possible. This leads to the equilibrium point (0, 0).
If x = 10 000, the second equation is
−0.1y(−39 + 0.1y) = 0,
which gives y = 0 or y = 390. So we have found two more equilibrium
points, namely (10 000, 0) and (10 000, 390).
So this system has three equilibrium points, namely (0, 0), (10 000, 0)
and (10 000, 390).
49
Unit 12 Systems of differential equations
(b) We have
u(x, y) = 0.5x − 0.000 05x2 ,
v(x, y) = −0.1y + 0.0004xy − 0.01y 2 .
So the Jacobian matrix is
% -
0.5 − 0.0001x 0
J(x, y) = .
0.0004y −0.1 + 0.0004x − 0.02y
(c) At the equilibrium point (0, 0),
% -
0.5 0
J(0, 0) = ,
0 −0.1
and the linearised approximations to the differential equations near
this equilibrium point are
% - % -% -
ṗ 0.5 0 p
= .
q̇ 0 −0.1 q
At the equilibrium point (10 000, 0),
% -
−0.5 0
J(10 000, 0) = ,
0 3.9
and the linearised approximations to the differential equations near
this equilibrium point are
% - % -% -
ṗ −0.5 0 p
= .
q̇ 0 3.9 q
Finally, at the equilibrium point (10 000, 390),
% -
−0.5 0
J(10 000, 390) = ,
0.156 −3.9
and the linearised approximations to the differential equations near
this equilibrium point are
% - % -% -
ṗ −0.5 0 p
= .
q̇ 0.156 −3.9 q
Solution to Exercise 13
(a) As the given matrix is lower triangular, the eigenvalues can be read off
the leading diagonal, so the eigenvalues are 1 and 3.
(b) As the eigenvalues are positive and distinct, the equilibrium point is a
source.
Solution to Exercise 14
(a) The characteristic equation of the matrix of coefficients is
−λ(−3 − λ) + 2 = 0,
which simplifies to
λ2 + 3λ + 2 = 0,
50
Solutions to exercises
Solution to Exercise 15
(a) The characteristic equation of the matrix of coefficients is
' '
'1 − λ 2 '
' ' = 0,
' 2 −2 − λ'
or λ2 + λ − 6 = 0, which factorises to give
(λ − 2)(λ + 3) = 0,
so the eigenvalues are λ = 2 and λ = −3.
The eigenvectors (a b)T corresponding to λ = 2 satisfy the equations
−a + 2b = 0,
2a − 4b = 0.
So an eigenvector corresponding to the positive eigenvalue λ = 2 is
(2 1)T , and all the eigenvectors are along the line q = 12 p.
The eigenvectors (a b)T corresponding to λ = −3 satisfy the
equations
4a + 2b = 0,
2a + b = 0.
So an eigenvector corresponding to the negative eigenvalue λ = −3 is
(1 −2)T , and all the eigenvectors are along the line q = −2p.
(b) The matrix of coefficients has a positive eigenvalue and a negative
eigenvalue, so the equilibrium point is a saddle.
(c) There are two straight-line paths, namely q = 12 p and q = −2p. On
the line q = 12 p, the point (p(t), q(t)) moves away from the origin as t
increases, because the corresponding eigenvalue is positive. On the line
q = −2p, the point approaches the origin as t increases, because the
corresponding eigenvalue is negative. This information, together with
the knowledge that the equilibrium point is a saddle, allows us to
sketch the phase portrait shown in the margin.
Solution to Exercise 16
(a) The characteristic equation of the matrix of coefficients is
(2 − λ)(−2 − λ) + 5 = 0,
that is, λ2 + 1 = 0, so the eigenvalues are λ = i and λ = −i.
(b) As the eigenvalues are imaginary, the equilibrium point is a centre.
51
Unit 12 Systems of differential equations
Solution to Exercise 17
(a) The characteristic equation of the matrix of coefficients is
(1 − λ)2 + 1 = 0,
that is, λ2 − 2λ + 2 = 0, which has complex roots λ = 1 + i and
λ = 1 − i.
(b) As the eigenvalues are complex with positive real part, the equilibrium
point is a spiral source.
Solution to Exercise 18
(a) As the matrix is diagonal, the eigenvalues can be read off the leading
diagonal. So the eigenvalue is 2 (repeated).
In fact, the matrix is twice the identity matrix, so any vector
transforms to twice itself. So any non-zero vector is an eigenvector. So
we can choose (1 0)T and (0 1)T to be two linearly independent
eigenvectors.
(b) The differential equations are
ṗ = 2p, q̇ = 2q,
which have general solution
p(t) = Ce2t , q(t) = De2t ,
where C and D are arbitrary constants.
(c) Eliminating t from the general solution, the equations of the paths are
D
q= p = Kp (C '= 0),
C
where K = D/C is also an arbitrary constant. So the paths are all
straight lines passing through the origin.
The above analysis has neglected the possibility C = 0. In this case
the path is p = 0, which is also a straight line passing through the
origin, namely the q-axis.
(d) Both p(t) and q(t) are increasing functions of time, so the point
(p(t), q(t)) moves away from the origin as t increases. So the
equilibrium point is unstable.
Solution to Exercise 19
(a) The characteristic equation is
(2 − λ)2 = 0,
so the matrix has the repeated eigenvalue λ = 2.
The eigenvectors (a b)T corresponding to this repeated eigenvalue
satisfy the equations
0 = 0, a = 0,
52
Solutions to exercises
Solution to Exercise 20
The characteristic equation of the matrix of coefficients is
λ2 + hk = 0.
√
Because h, k are positive, the eigenvalues are λ = ±i hk, so the
equilibrium point is a centre.
Solution to Exercise 21
The eigenvalues of the matrix of coefficients are λ = k and λ = −h, where
h, k are positive. So the equilibrium point is a saddle. (In fact, in this case
we have to restrict p and q to non-negative values, but this does not affect
our conclusion.)
Solution to Exercise 22
(a) The characteristic equation is
λ2 − 4λ + 13 = 0,
so the eigenvalues are 2 + 3i and 2 − 3i, corresponding to the
eigenvectors (1 −i)T and (1 i)T , respectively.
As the eigenvalues are complex with a positive real component, the
equilibrium point is a spiral source.
(b) As the equilibrium point of the linear approximation is not a centre,
the corresponding equilibrium point of the non-linear system is also a
spiral source.
53
Unit 12 Systems of differential equations
Solution to Exercise 23
(a) The equilibrium points are given by
(1 + x − 2y)x = 0,
(x − 1)y = 0.
The second equation gives
x=1 or y = 0.
When x = 1, substituting into the first equation gives
2 − 2y = 0,
which leads to y = 1. So (1, 1) is an equilibrium point.
When y = 0, substituting into the first equation gives
(1 + x)x = 0,
hence x = 0 or x = −1. So we have found two further equilibrium
points, namely (0, 0) and (−1, 0).
Thus we have three equilibrium points: (1, 1), (0, 0) and (−1, 0).
(b) With the usual notation,
u(x, y) = (1 + x − 2y)x = x + x2 − 2xy,
v(x, y) = (x − 1)y = xy − y.
So the Jacobian matrix is
% - % -
ux uy 1 + 2x − 2y −2x
= .
vx vy y x−1
(c) At the point (0, 0), the Jacobian matrix is
% -
1 0
,
0 −1
so the linearised system is
% - % -% -
ṗ 1 0 p
= .
q̇ 0 −1 q
The eigenvalues of the matrix of coefficients are λ = 1 and λ = −1. As
one of the eigenvalues is positive and the other is negative, the
equilibrium point of the linearised system is a saddle.
At the point (−1, 0), the Jacobian matrix is
% -
−1 2
,
0 −2
so the linearised system is
% - % -% -
ṗ −1 2 p
= .
q̇ 0 −2 q
54
Solutions to exercises
Solution to Exercise 24
(a) To answer this part of the exercise we use the equation
* y)
ẋ = x 1 − .
2
As this equation is already factorised, we deduce that ẋ = 0 when
x = 0 or when y = 2.
For ẋ to be positive, either both terms in the product must be positive
or both terms must be negative. Both terms are positive when x > 0
and y < 2. Both terms are negative when x < 0 and y > 2. This gives
two regions where the growth rate ẋ is positive.
The figure in the margin is a sketch of these lines and regions. The
two lines where ẋ is zero are marked, and the two regions where ẋ is
positive are shaded.
55
Unit 12 Systems of differential equations
Solution to Exercise 25
(a) The equilibrium points are marked in the figure in the margin.
As described in Example 8, it is easiest to defer marking the paths
near equilibrium until after the nullcline crossings have been
determined. In this case the nullcline crossings indicate whether the
spiral source at (1, 1) is a clockwise or anticlockwise spiral.
(b) The nullclines for ẋ are x = 0 and 1 + x − 2y = 0 (which is the line
y = (1 + x)/2). These are shown as the two red lines in the figure in
the margin.
The nullclines for ẏ are y = 0 and x = 1, which are shown as the two
blue lines in the figure in the margin.
(c) The nullcline x = 0 has one equilibrium point on it, so to determine
the direction of the crossings, we evaluate ẏ at two points.
(x, y) (0, 1) (0, −1)
ẏ −1 1
The signs of these values are marked by the red arrows along the
y-axis in the figure in the margin.
The nullcline y = (1 + x)/2 has two equilibrium points on it, so to
determine the direction of the crossings, we evaluate ẏ at three points.
(x, y) (−2, −1/2) (0, 1/2) (3, 2)
ẏ 3/2 −1/2 4
These are represented by up and down arrows marked in the figure in
the margin.
The nullcline y = 0 has two equilibrium points on it, so we need to
evaluate ẋ at three points.
(x, y) (−2, 0) (−1/2, 0) (1, 0)
ẋ 2 −1/4 2
These are represented by left and right arrows marked in the figure in
the margin.
56
Solutions to exercises
57
Unit 13
Fourier series
Introduction
Introduction
In Unit 7 you saw that many functions can be approximated by a Taylor
polynomial
1
f (x) * f (x0 ) + f ! (x0 ) (x − x0 ) + f !! (x0 ) (x − x0 )2 + · · ·
2!
1 (n) n
+ f (x0 ) (x − x0 ) .
n!
It is often the case that a small number of terms gives a useful
approximation, and it is tempting to ask whether the approximation may
be made exact by taking an infinite number of terms – in other words, is
#∞
1 (n)
f (x) = f (x0 ) (x − x0 )n
n=0
n!
true? This is indeed true for sufficiently smooth functions, but it is not
necessarily true for all functions. An example for which the above formula
is true is when f (x) = exp(x) and x0 = 0. In this case we have
f (n) (x0 ) = e0 = 1, so
∞
# 1 n
ex = x ,
n!
n=0
for any real number x.
In this unit we are primarily concerned not with polynomial functions
(which, if not constant, become numerically very large as x → ±∞), but
with periodic functions, such as sin x and cos x. A great deal of beautiful
mathematics has arisen from the analysis that we will describe, and there
are also important practical benefits.
One example arises directly from Unit 10, where you studied differential
equations modelling forced and damped oscillations. You saw how they
could be used to predict the response of various mechanical and electrical
systems. In particular, you saw how they responded to a sinusoidal forcing
term like cos t, with graph as in Figure 1.
61
Unit 13 Fourier series
The graph in Figure 2 is similar to that of cos t, except that all the values
are positive. The function b(t) is described by
b(t) = |cos t|.
62
Introduction
This graph differs from that of the cosine function, which turns smoothly.
Here the direction changes abruptly every time the graph reaches the
t-axis. However, b(t) is still continuous in that there are no sudden jumps
in the function value as t increases smoothly; it is possible to draw the
graph without taking the pen off the paper.
The graph of g(t) in Figure 3 is also continuous, but with abrupt changes
of direction at both the highest and lowest values. Between the points
where the direction changes, the graph is a straight line. The whole graph
looks rather like the blade of a saw, so g(t) is known as a sawtooth
function. You will meet this function again in Unit 14, where it will be
used to model the initial displacement of a plucked guitar string.
The graph in Figure 4 shows a function that takes the value 1 whenever t
lies in the interval [(2k − 21 )π, (2k + 12 )π] for some integer k, and the
value 0 otherwise, that is,
'
1 for (2k − 21 )π ≤ t ≤ (2k + 12 )π (k ∈ Z),
h(t) =
0 otherwise.
This graph differs quite radically from those of cos t, b(t) and g(t). Here
there are abrupt jumps in the function value itself (rather than merely
abrupt changes of direction) at the points t = (k + 21 )π. The function h(t)
is discontinuous, while cos t, b(t) and g(t) are continuous. The function
h(t) is known as a square-wave function.
At present we have no way of predicting the response of a mechanical or
electrical system to a forcing function like b(t), g(t) or h(t) (although such
systems are extremely common). Fourier series provide the answer. The
differential equation
mẍ + rẋ + kx = P cos(Ωt)
is linear. Therefore the principle of superposition tells us that if See Unit 1.
x = P1 M1 cos(Ω1 t + φ1 )
is a solution of
mẍ + rẋ + kx = P1 cos(Ω1 t),
and
x = P2 M2 cos(Ω2 t + φ2 )
is a solution of
mẍ + rẋ + kx = P2 cos(Ω2 t),
then
x = P1 M1 cos(Ω1 t + φ1 ) + P2 M2 cos(Ω2 t + φ2 )
is a solution of
mẍ + rẋ + kx = P1 cos(Ω1 t) + P2 cos(Ω2 t).
63
Unit 13 Fourier series
64
1 Introducing Fourier series
Exercise 1
Let G(t) be defined as in equation (4). Find G(t + 2π) in terms of G(t).
You saw in Exercise 1 that the function G(t) defined by equation (4) is
periodic with period 2π, just like the cosine function cos t. But what of the
individual terms in the sum? Apart from the constant term, they are
multiples of
cos t, cos 3t, cos 5t, cos 7t, ... . (5)
65
Unit 13 Fourier series
Exercise 2
What are the angular frequencies and periods of the functions in
sequence (5)?
(Recall that in the expression cos(ωt), the constant ω is called the angular
frequency. The angular frequency is related to the period using the fact
that the cosine function will repeat when its argument increases by 2π, so
ωτ = 2π, where τ is the period, that is, τ = 2π/ω.)
From Exercise 2 you can see that we have a family of cosine functions
where all the angular frequencies are integer multiples of the smallest
angular frequency 1, and whose periods are integer fractions of the
fundamental period 2π. We have seen that this is the period of the
function G(t), since all the component functions will have repeated after
this time – some having repeated several times.
More generally (as you may recall from Unit 9), any function f (t) is said
to be periodic if it repeats regularly, that is, if there is some positive
value λ such that for all t, f (t + λ) = f (t). In this case, it is also true that
for all t, f (t + 2λ) = f (t + λ) = f (t), so 2λ could be taken as a period for
f (t) instead of λ, and in general, nλ could be taken as the period (where n
is any positive integer). The fundamental period of a periodic function
is the smallest possible (positive) value for the period.
In applications, the fundamental period is far more important than the
other periods. For this reason, ‘fundamental period’ is usually shortened to
simply period. For example, the fundamental period of a pendulum (the
time that it takes to swing to and fro) is usually called the period of the
pendulum. We will occasionally use this shorthand when there is no risk of
confusion. If we talk about the period of a function, then we mean its
fundamental period.
In this unit, time is the independent variable of functions, and τ is used to
denote the period of a periodic function.
Example 1
Let f (t) = cos 4t + 3 cos 6t. What are the angular frequencies and
corresponding periods of the component functions? What is the period of
the function f (t)?
Solution
The angular frequencies of the component functions are 4 and 6. Their
corresponding periods are π2 and π3 , respectively. Hence the period of the
combined function f (t) is τ = π (as this is the shortest time that is an
integer multiple of both π2 and π3 ). After this time, the first cosine term
will have completed two cycles, while the other cosine term will have
completed three.
66
1 Introducing Fourier series
Exercise 3
Let f (t) = 2 cos πt + 3 cos 3π
2 t − cos 2πt. What are the angular frequencies
and corresponding periods of the component functions? What is the
period of the function f (t)?
Exercise 4
Sketch the
* graphs ( of the functions C2 (t) and C3 (t) on the fundamental
interval − 2τ , 2τ .
What happens if you try to define C0 (t) using formula (6)?
67
Unit 13 Fourier series
Both graphs exhibit symmetry. You can see that the graph of the cosine
function C1 (t) takes the same values at corresponding points on either side
of the vertical axis. We say that the function is even. By contrast, in the
graph of the sine function S1 (t), the values at corresponding points on
either side of the vertical axis have the same magnitude but opposite signs.
We say that the function is odd.
68
1 Introducing Fourier series
Example 2
Suppose that the function f (t) is defined by f (t) = t2 . Is this function
even, odd, or neither even nor odd?
Solution
Since f (−t) = (−t)2 = t2 = f (t) for all t, the function is even.
Exercise 5
Suppose that the function g(t) is defined by g(t) = t3 . Is this function
even, odd, or neither even nor odd?
Exercise 6
If f (t) and g(t) are both odd functions, show that the function k(t) defined
by
k(t) = f (t) + g(t)
is also an odd function.
The graphs of the functions f (t) and g(t) defined in Example 2 and
Exercise 5, and shown in Figure 8, should make the definitions clearer.
Figure 8 Graphs of the even function f (t) = t2 and the odd function
g(t) = t3
For the even function f (t), the same values appear on either side of the
vertical axis, so the graph has reflection symmetry about this line. For the
odd function g(t), the values on either side of the vertical axis have
opposite signs, so the graph has rotational symmetry through the angle π
about the origin.
Generalising from the function f (t) = t2 , we can state that polynomial
functions where all the powers are even are themselves even functions.
Similarly, polynomial functions where all the powers are odd are
themselves odd functions. Indeed, this is the origin of the terms
‘even function’ and ‘odd function’.
69
Unit 13 Fourier series
Exercise 7
(a) Is the function C0 (t), given by C0 (t) = 1 for −τ/2 ≤ t ≤ τ/2, even, odd
or neither?
(b) Is the function h(t), defined by h(t) = t2 + t3 for all t, even, odd or
neither?
The way that even and odd functions combine is similar to the way that
positive and negative numbers combine. That is:
• the sum of two even functions (positive numbers) is even (positive)
• the sum of two odd functions (negative numbers) is odd (negative)
• the sum of an even function (positive number) and an odd function
(negative number) is neither even nor odd (positive, negative or zero)
• the product of two even functions (positive numbers) is even (positive)
• the product of two odd functions (negative numbers) is even (positive)
• the product of an even function (positive number) and an odd
function (negative number) is odd (negative).
In the next example and exercise, we use the first two properties in the list
above to demonstrate the last two properties.
Example 3
If f (t) = t3 + 2t5 and g(t) = t − t3 , show that the function defined by
h(t) = f (t) g(t) is an even function.
Solution
Calculating explicitly,
h(t) = f (t) g(t) = (t3 + 2t5 )(t − t3 )
= t4 − t6 + 2t6 − 2t8
= t4 + t6 − 2t8 .
This is a polynomial where all the powers are even, therefore h(t) is an
even function.
Alternatively, since both f (t) and g(t) are odd functions, we know by
definition that
f (−t) = −f (t), g(−t) = −g(t).
Hence
h(−t) = f (−t) g(−t) = (−f (t))(−g(t)) = f (t) g(t) = h(t),
so h(t) is an even function.
70
1 Introducing Fourier series
Exercise 8
If f (t) = t3 + 2t5 and g(t) = 3t2 − t4 , show that the function defined by
h(t) = f (t) g(t) is an odd function.
For much of this subsection we have been concerned with even and odd
functions that are not periodic. If a function f (t) is periodic, of period 2a,
then an advantage of choosing a fundamental interval [−a, a] centred on
the origin is that we can tell whether f (t) is even, odd or neither by seeing
whether it is even, odd or neither on [−a, a].
Now we investigate the properties of odd and even functions that will
simplify later calculations, namely what happens when these functions are
integrated over the fundamental interval [−a, a].
For an odd function f (t), the integral is zero because the integral for
positive t-values is exactly cancelled by the integral for negative t-values.
This is illustrated in Figure 9 and is proved by the following argument.
Calling the integral I and splitting it into two halves gives
! a ! 0 ! a
I= f (t) dt = f (t) dt + f (t) dt.
−a −a 0
Figure 9 An odd function
Now use the rule that changing the order of integration changes the sign of with area above the
the integral to obtain horizontal axis shaded blue
! −a ! a and area below shaded red.
I=− f (t) dt + f (t) dt. By symmetry, the two areas
0 0 are equal in size, so the
integral is zero (as the red
We can use the substitution u = −t to yield
! a ! a area is counted as negative).
I= f (−u) du + f (t) dt.
0 0
Finally, use the fact that f is odd, so f (−u) = −f (u), to get
! a ! a
I=− f (u) du + f (t) dt = 0,
0 0
where the final equality follows because the two integrals are the same
integral written with different integration variables.
If f is an even function, then the argument above can be applied up until
the penultimate step, then instead of terms cancelling, the result will be
twice the integral from 0 to a.
These results are worth remembering as they can save a lot of effort when
evaluating integrals of even and odd functions.
71
Unit 13 Fourier series
From Figure 7, you can see that the cosine function is even, but the sine
function is odd. The suggestion is that to approximate an even function
(such as the sawtooth function or the square-wave function) as a sum of
sinusoidal terms, we ought to ensure that the approximating function is
even. The only way to do this is to ensure that only cosine functions
appear in the sum, as you will see in the next section.
72
2 Fourier series for even functions with period 2π
Exercise 9
What are the period and angular frequency of the sawtooth function g(t)?
73
Unit 13 Fourier series
Exercise 10
Verify that the integral in formula (12) is zero (for each n = 1, 2, 3, . . .).
In this module we assume the If we integrate both sides of equation (11) term by term over the
validity of term-by-term fundamental interval [−π, π], then we find that
integration and differentiation of
! π ! π2 #∞
4
infinite series. The process is
valid in all the practical cases F (t) dt = A0 + An cos nt dt
that concern us. −π −π n=1
! π ∞
# ! π
= A0 dt + An cos nt dt.
−π n=1 −π
We do not know the coefficients of F (t), but our aim is to ensure that
F (t) = f (t). So to determine A0 , we replace F (t) by f (t) in the above
equation to obtain the following result.
! π
1
A0 = f (t) dt. (13)
2π −π
You can think of A0 as the average value taken by the function f (t).
Example 4
Find the value of A0 when the general function f (t) is replaced by the
sawtooth function g(t) given by equation (9).
74
2 Fourier series for even functions with period 2π
Solution
Equation (13) becomes
! π
1
A0 = g(t) dt.
2π −π
As the function g(t) is even, equation (7) applies and the integral is equal
to twice the integral over the positive t-values:
!
1 π
A0 = g(t) dt.
π 0
Using this fact simplifies the calculation as g(t) is defined piecewise, with
different formulas for positive and negative values; here we need to
consider only one of the formulas.
The function g(t) is given by equation (9), so the above integral is
! ! . / ) 5π
1 π 1 π 1 1 1 2 1
g(t) dt = − t + 1 dt = − t +t = .
π 0 π 0 π π 2π 0 2
Thus for the case of the sawtooth function,
A0 = 21 .
Exercise 11
Suppose that the general function f (t) is now replaced by the square-wave
function h(t) of the Introduction, which can be defined on the fundamental
interval [−π, π] by
'
1 for − π2 ≤ t ≤ π2 ,
h(t) =
0 otherwise.
Find the value of the Fourier coefficient A0 .
Hint: As the function h(t) is zero over some of the fundamental interval,
! π ! π/2
h(t) dt = h(t) dt.
−π −π/2
Exercise 12
(a) Show that
! π
A0 cos t dt = 0.
−π
(c) More generally, use the identity given in part (b) to show that
! π
cos nt cos t dt = 0 when n is an integer and n > 1.
−π
76
2 Fourier series for even functions with period 2π
! π
1
A1 = f (t) cos t dt. (16)
π −π
Both of these integrals are easy to derive using integration by parts, but it
is quicker to state the standard result. One of these integrals will be used
in the calculation of A1 for the sawtooth function in the next example.
Example 5
Returning to the sawtooth function g(t) defined by equation (9), find the
value of the coefficient A1 .
Solution
The coefficient is given by
!
1 π
A1 = g(t) cos t dt.
π −π
As g(t) is even and cos t is even, the product g(t) cos t is even, and we may
make use of equation (7) to simplify the calculation by evaluating twice
the integral over the positive t-values:
!
2 π
A1 = g(t) cos t dt.
π 0
Substituting the definition of g(t) from equation (9) gives
! . /
2 π 1
A1 = − t + 1 cos t dt
π 0 π
! π !
2 2 π
=− 2 t cos t dt + cos t dt.
π 0 π 0
77
Unit 13 Fourier series
The first integral is one of the two useful integrals (equation (18) with
a = 1) that were stated immediately preceding this example, so performing
the integrations yields
2* (π 2 * (π
A1 = − 2 cos t + t sin t 0 + sin t 0
π π
2 4
= − 2 (−1 − 1) = 2 .
π π
78
2 Fourier series for even functions with period 2π
and using formulas (19) and (20), we find that all of the terms of the above
infinite sum are zero except when n = m, so we get
! π ! π
F (t) cos mt dt = Am cos2 mt dt.
−π −π
! π
1
An = f (t) cos nt dt (n > 0). (22)
π −π
Example 6
Returning once again to the sawtooth function g(t) as defined by
equation (9), find the values of the coefficients A2 and A3 .
Solution
Substituting g(t) into equation (22) gives
!
1 π
An = g(t) cos nt dt.
π −π
Now we use the fact that the integrand is even, since it is the product of
two even functions, g(t) and cos nt, to obtain
! . /
2 π 1
An = − t + 1 cos nt dt
π 0 π
! π !
2 2 π
=− 2 t cos nt dt + cos nt dt,
π 0 π 0
where we have substituted for g(t) using the definition in equation (9).
79
Unit 13 Fourier series
Exercise 14
Suppose that f (t) in formula (22) is replaced by the square-wave function
h(t), defined in Exercise 11. Find the values of the coefficients A2 and A3 .
80
2 Fourier series for even functions with period 2π
Exercise 15
You have found, in Exercises 11, 13 and 14, the Fourier coefficients A0 , A1 ,
A2 and A3 for the square-wave function h(t) defined in Exercise 11. In
fact, as indicated in equation (3), its Fourier series is
1 2 2 2 2
H(t) = + cos t − cos 3t + cos 5t − cos 7t + · · · .
2 π 3π 5π 7π
Write down this series in closed form.
The following exercises ask you to apply the formulas derived in this
section to find Fourier series for other even periodic functions.
Exercise 16
Find the Fourier series for the even periodic function f (t) defined on the
fundamental interval [−π, π] by
f (t) = t2 .
(Hint: You may find the following integral obtained by integration by
parts useful:
!
1 & -
t2 cos nt dt = 3 (n2 t2 − 2) sin nt + 2nt cos nt + C,
n
where C is a constant.)
Exercise 17
A variant of the sawtooth function can be defined on the fundamental
interval [−π, π] by
w(t) = |t|.
Recall that the absolute value function is defined by
'
t for t ≥ 0,
|t| =
−t for t < 0.
Find the Fourier series for this function, and write down the
approximation W5 (t).
81
Unit 13 Fourier series
2.5 Convergence
You have just tackled a substantial piece of work, and this has involved
finding Fourier series for several even functions. Having found them, you
need to take stock of what you have done. You have seen that just a few
terms of the Fourier series for the sawtooth function g(t) give a very good
approximation. However, the first few terms of the Fourier series for the
square-wave function h(t) do not give a particularly good approximation,
as Figure 14 illustrates.
The situation improves only slowly for the square-wave function. Plotting
the graphs for the sums as far as the cos 7t, cos 11t and cos 21t terms,
better approximations to h(t) are obtained, as expected (see Figure 15).
However, even H21 (t) does not approximate h(t) as well as G3 (t) does g(t)
in Figure 13. This is because of the discontinuities in h(t). We cannot
reasonably expect the sum of continuous sinusoidal functions to provide a
good approximation to a discontinuous function. From the graphs, you can
see that the approximations to h(t) are worse near the discontinuities, that
is, near the points where the value of h(t) jumps from 0 to 1 and back
again. Nevertheless, even for a discontinuous function such as h(t), we can,
remarkably, approximate reasonably well using Fourier series. At a
discontinuity, the Fourier series takes the average value of the function at
either side of the discontinuity. This is formally stated in the following
theorem (which we do not prove) that guarantees the nature of the Fourier
series for a wide class of functions at points in the fundamental interval.
82
2 Fourier series for even functions with period 2π
Example 7
Consider the function
'
−1 for −1 ≤ t < 0,
f (t) =
t for 0 ≤ t < 1,
f (t + 2) = f (t),
and its corresponding Fourier series F (t).
(a) Calculate f (−1), f ( 12 ) and f (2).
(b) Calculate F (−1), F ( 12 ) and F (2).
(c) Compare the values obtained in parts (a) and (b).
Solution
(a) The first step in solving problems such as this is to draw a sketch
graph, such as the one shown in Figure 16.
Using the sketch as a guide, we can calculate
f (−1) = −1,
f ( 12 ) = 12 ,
f (2) = f (0) = 0. Figure 16 Sketch graph of
(b) The pointwise convergence theorem gives f (t) on its fundamental
interval
f (−1+ ) + f (−1− )
F (−1) = .
2
Approaching t = −1 from above is included in the fundamental
interval of f , so we can say that f (−1+ ) = −1.
83
Unit 13 Fourier series
Exercise 18
Consider the function
|t| + t
f (t) = for −1 ≤ t < 1,
2
f (t + 2) = f (t),
and its corresponding Fourier series F (t). Calculate the values F (−1)
and F (0).
84
3 Fourier series for even and odd periodic functions
Figure 17 Graph of the function w(t) over four periods, with the
fundamental interval highlighted in red
85
Unit 13 Fourier series
Since we are dealing with even functions, we also include the constant
function
C0 (t) = 1.
As in Section 2, the argument used here is a general one. Thus for the
remainder of this subsection, we use the symbol f*(t) to (refer to a general
even periodic function with fundamental interval − 2τ , 2τ . As before, we
assume that we can choose the coefficients in an infinite sum of these
functions Cn (t) in such a way that we can approximate the original
function as accurately as required. We write this (as before) as
#∞ . /
2nπt
F (t) = A0 + An cos . (24)
n=1
τ
To find the coefficients in this sum, we need to evaluate integrals of the
cosine functions that generalise equations (19)–(21) used in the previous
section. The integrals in which* we are( interested are for functions defined
over the fundamental interval − 2τ , 2τ , which are obtained from the former
integrals by the substitution u = τ t/(2π).
* (
Trigonometric integrals over the interval − 2τ , 2τ
For any positive integers m and n,
! τ/2 . /
2mπt
cos dt = 0, (25)
−τ/2 τ
! τ/2 . / . /
2mπt 2nπt
cos cos dt = 0 (m += n), (26)
−τ/2 τ τ
! τ/2 . /
2 2mπt τ
cos dt = . (27)
−τ/2 τ 2
Using formula (25), all the integrals in the infinite sum become zero,
leaving
! τ/2 ! τ/2
F (t) dt = A0 dt = τA0 ,
−τ/2 −τ/2
86
3 Fourier series for even and odd periodic functions
so
! τ/2
1
A0 = F (t) dt.
τ −τ/2
As in Section 2, we now use the fact that we wish to choose the coefficients
so that F (t) = f (t). Thus we put F (t) = f (t) to give the following result.
! τ/2
1
A0 = f (t) dt. (28)
τ −τ/2
! τ/2 . /
2 2nπt
An = f (t) cos dt. (29)
τ −τ/2 τ
Exercise 19
For the sawtooth function w(t) defined by equation (23), find the
coefficients A0 and An .
87
Unit 13 Fourier series
Figure 18 The sawtooth function w(t) compared to the first two Fourier
series approximations
88
3 Fourier series for even and odd periodic functions
Example 8
Sketch the graphs of the
* functions
( S1 (t), S2 (t), S3 (t) and S4 (t) on the
fundamental interval − 2τ , 2τ .
Solution
Sketches of the four functions are shown in Figure 20.
As before, the argument used to find the coefficients in a Fourier series for
an odd function is a general one. Thus for the remainder of this subsection
we use the function f *(t) to refer
( to a general odd periodic function with
fundamental interval − 2τ , 2τ . Then we assume that we can choose the
coefficients in an infinite sum of the functions Sn (t) in such a way that we
can approximate the original function f (t) as accurately as required.
89
Unit 13 Fourier series
* (
Trigonometric integrals over the interval − 2τ , 2τ
For any positive integers m and n,
! τ/2 . /
2mπt
sin dt = 0, (34)
−τ/2 τ
! τ/2 . / . /
2mπt 2nπt
sin sin dt = 0 (m += n), (35)
−τ/2 τ τ
! τ/2 . /
2 2mπt τ
sin dt = . (36)
−τ/2 τ 2
Using formula (35), all the integrals on the right-hand side of equation (33)
become zero except when n = m. That is,
! τ/2 . / ! τ/2 . /
2mπt 2mπt
F (t) sin dt = Bm sin2 dt.
−τ/2 τ −τ/2 τ
Using formula (36), the right-hand side reduces to Bm τ/2, and we can
make Bm the subject (and as before replace F (t) by the function f (t) and
rewrite the index as n instead of m).
! τ/2 . /
2 2nπt
Bn = f (t) sin dt. (37)
τ −τ/2 τ
90
3 Fourier series for even and odd periodic functions
Example 9
(a) Find the coefficients B1 , B2 and B3 for the function v(t) defined in
equation (31).
(b) Sketch the graph of
. / . / . /
2πt 4πt 6πt
V3 (t) = B1 sin + B2 sin + B3 sin
τ τ τ
* τ τ(
on the interval − 2 , 2 , and compare it with the graph of v(t).
Solution
(a) The coefficients Bn are given by formula (37) with f (t) = t:
! . /
2 τ/2 2nπt
Bn = t sin dt.
τ −τ/2 τ
As the function f (t) = t is odd and the sine term is odd, the product
is even, so we can use equation (7) to rewrite the integral as twice the
sum over the positive t-values:
! . /
4 τ/2 2nπt
Bn = t sin dt.
τ 0 τ
This integral is one of the two useful integrals (equation (17) with
a = 2nπ/τ), so
) 2 . . / . //5τ/2
4 τ 2nπt 2nπt 2nπt
Bn = sin − cos
τ 4n2 π2 τ τ τ 0
) . / . /5τ/2
τ 2nπt 2nπt 2nπt
= 2 2 sin − cos
n π τ τ τ 0
τ
= 2 2 (−nπ cos nπ)
n π
τ(−1)n
=− .
nπ
So B1 = τ/π, B2 = −τ/(2π) and B3 = τ/(3π).
(b) Substituting the coefficients into the given equation and simplifying
gives
) . / . / . /5
τ 2πt 1 4πt 1 6πt
V3 (t) = sin − sin + sin .
π τ 2 τ 3 τ
* (
The graphs of v(t) and V3 (t) on − 2τ , 2τ are shown in Figure 21.
91
Unit 13 Fourier series
Our sketch in Figure 21 of the sum of the first three terms gives an
approximation to the function, but it is not as good as the corresponding
approximation to the sawtooth function that you met in the previous
subsection. As in the case of the square-wave function studied in
Subsection 2.5, this is due to the discontinuities in the original function.
At these points, the Fourier series takes the average value in the middle of
the jump. Here that value is 0. In the case of a continuous function, the
sizes of the coefficients in the Fourier series generally have an n2 factor in
the denominator and so decrease quite rapidly (such as in equation (30)).
By contrast, here and for the square-wave function, where there are
discontinuities, the sizes of the coefficients of the Fourier series have only a
factor n in the denominator and so the coefficients decrease more slowly.
This is a general phenomenon: functions with discontinuities converge
more slowly than smooth functions.
Now try to find a Fourier series for an odd function yourself by working
through the following exercise.
Exercise 20
The periodic function f (t) with period τ is defined on the interval [− 2τ , 2τ ]
by
'
−1 for − 2τ < t ≤ 0,
f (t) =
1 for 0 < t < 2τ .
Find the first three non-zero terms of the Fourier series for this function.
Exercise 21
Consider a general function f (x).
(a) Show that the function g(x) defined by
f (x) + f (−x)
g(x) =
2
is even.
92
4 Fourier series for any periodic function
Exercise 21 shows that one way of finding the Fourier series for a general
function f is to find Fourier series for the functions g and h as defined in
the exercise, and then add them. However, we can find the Fourier series
for a general function more directly, as you will see in Subsection 4.1.
The modelling of a real problem may involve a function f (t) that is defined
* τ(
only on some interval. We can choose the interval to be of the form* 0, 2( .
Then we can extend the definition of the function to the interval − 2τ , 2τ
by choosing the function to be either even or odd on this interval. From
there, we can extend the definition of the function to all the real numbers
as a periodic function. You will see how to do this in Subsection 4.2.
Using this result, no new terms appear when we form our products, so we
arrive at the same formulas as before, which are summarised as follows.
93
Unit 13 Fourier series
It should be noted that the integrals for determining A0 , An and Bn are all
over intervals that are symmetric about the origin, which means that the
results for integrals of even and odd functions (equations (7) and (8)) will
apply. In particular, if f is even then Bn is zero, and if f is odd then A0
and An are zero.
The Fourier coefficients of a given function are unique. So if a function is
itself a sum of sine and cosine functions, then it is its own Fourier series.
For example, the Fourier series for 12 sin 3t is 21 sin 3t – that is, B3 = 12 and
all other Fourier coefficients are zero.
Example 10
The periodic function f (t) is defined by
f (t) = et for −1 ≤ t ≤ 1,
on the fundamental interval [−1, 1]. Find its Fourier series.
Solution
The function f (t) has the graph shown in Figure 22.
This function is clearly neither even nor odd. Using Procedure 1, with
period τ = 2, we first obtain
! 1
A0 = 21
et dt = 12 (e − e−1 )
−1
and
! 1
An = et cos(nπt) dt.
−1
but this is easier as this integral has already appeared in equation (39)
during the computation of An . Substituting for An in equation (39) gives
(e − e−1 )(−1)n
= (e − e−1 )(−1)n + nπBn .
1 + n2 π2
Rearranging this equation and taking out common factors gives
. /
(e − e−1 )(−1)n 1
Bn = −1 .
nπ 1 + n2 π2
Simplifying then gives
nπ(e − e−1 )(−1)n
Bn = − .
1 + n2 π2
95
Unit 13 Fourier series
Every coefficient in the Fourier series involves the term e − e−1 . The series
is therefore conveniently written as
#∞ . / # ∞ . /
2nπt 2nπt
F (t) = A0 + An cos + Bn sin
τ τ
n=1 n=1
2 ∞
4
1 # (−1) n & -
= (e − e−1 ) + cos(nπt) − nπ sin(nπt) .
2 1 + n2 π2
n=1
The graph of the function f (t) and the graph of F8 (t), the sum of the
constant and the first eight cosine terms and first eight sine terms in the
Fourier series, are shown in Figure 23.
Exercise 22
Suppose that the periodic function f (t) is defined on the fundamental
interval [−1, 1] by
'
1 for −1 ≤ t < 0,
f (t) =
t for 0 ≤ t ≤ 1.
Find the coefficients of its Fourier series.
Exercise 23
(a) Find a fundamental interval for, and hence the period of, the function
b(t) defined in the Introduction as
b(t) = |cos t|.
96
4 Fourier series for any periodic function
(b) State whether b(t) is even, odd, or neither even nor odd, and find the
Fourier series for this function.
(Hint: Use the trigonometric identity
& -
cos t cos 2nt = 21 cos(2n − 1)t + cos(2n + 1)t ,
which is based on a more general identity given in the Handbook.)
97
Unit 13 Fourier series
The definition of the odd periodic extension needs a little more care so
that the resulting function is both periodic and odd. The reason for this is
that any function that is both periodic and odd is zero at the origin and at
the endpoints of a fundamental interval centred on the origin.
To show that any odd function f (t) has the value zero at the origin, let a
be the value at the origin, that is, a = f (0). Then since f is odd, we have
f (−0) = −f (0), thus a = −a, that is, 2a = 0 and so a = 0. To show that
any odd function with period 2T is zero at the right-hand endpoint, let
a = f (T ). As f is periodic with period 2T , we must have
f (−T ) = f (2T − T ) = f (T ) = a. As f is odd, we must have
f (−T ) = −f (T ), so this again leads to the equation a = −a, and as before
a = 0. These results are built into the definition of the odd extension of a
function in the following definition.
98
4 Fourier series for any periodic function
Example 11
Find and sketch the even and odd periodic extensions of the function
f (t) = t for 0 ≤ t ≤ 1.
Solution
The even periodic extension is given by
'
t for 0 ≤ t < 1,
feven (t) =
−t for −1 ≤ t < 0,
feven (t + 2) = feven (t).
This function is sketched in Figure 28, with the original function shown in
red.
Figure 28 The given function is shown in red and the even periodic
extension is shown in black
99
Unit 13 Fourier series
This function is sketched in Figure 29, with the original function shown in
red.
Figure 29 The given function is shown in red and the odd periodic
extension is shown in black
Exercise 24
Consider the function shown in Figure 30 and defined by
+
t for 0 ≤ t ≤ 1,
q(t) = 3 1
2 − 2 t for 1 < t ≤ 3.
Define the even and odd periodic extensions of q(t), simplifying the
Figure 30 formulas if possible. State the fundamental periods, and sketch each
extension over a range of three periods.
Example 12
Consider the function defined on the finite interval 0 ≤ t ≤ T by
'
t/T for 0 ≤ t < T /2,
f (t) =
(T − t)/T for T /2 ≤ t ≤ T,
where T is a positive constant. Express f (t) as a Fourier series that
involves only sine terms.
100
4 Fourier series for any periodic function
Solution
Because we are looking for a Fourier series that involves only sine terms,
we need to consider the odd periodic extension of f (t), denoted by fodd (t).
This is sketched in Figure 31.
The function fodd (t) is odd and has period τ = 2T , so its Fourier series
(from equation (32)) takes the form
∞
# . /
nπt
Fodd (t) = Bn sin ,
T
n=1
where the Fourier coefficients Bn are given by
! . /
1 T nπt
Bn = fodd (t) sin dt.
T −T T
The integrand is even as it is the product of two odd functions, so the
integral can be written as twice the integral over positive values. Also,
fodd (t) = f (t) on this interval, so
! . /
2 T nπt
Bn = f (t) sin dt.
T 0 T
Using the given piecewise definition of f (t), we obtain
! T /2 . / ! T . /
2 nπt 2 nπt
Bn = 2 t sin dt + 2 (T − t) sin dt.
T 0 T T T /2 T
Expanding the second integral gives
! T /2 . / ! . /
2 nπt 2 T nπt
Bn = 2 t sin dt + sin dt
T 0 T T T /2 T
! T . /
2 nπt
− 2 t sin dt.
T T /2 T
The first and third integrals are of the form of one of the two useful
integrals (equation (17) with a = nπ/T ), so we get
) 2 . . / . //5T /2
2 T nπt nπt nπt
Bn = 2 sin − cos
T n2 π2 T T T 0
) . /5T
2 T nπt
+ − cos
T nπ T T /2
) 2 . . / . //5T
2 T nπt nπt nπt
− 2 sin − cos .
T n2 π2 T T T T /2
101
Unit 13 Fourier series
Simplifying,
2 0 nπ nπ nπ $ 2 0 nπ $
Bn = sin − cos − cos nπ − cos
n2 π2 2 2 2 nπ 2
2 0 0 nπ nπ nπ $$
− 2 2 −nπ cos nπ − sin − cos .
n π 2 2 2
There is a lot of cancellation of terms, and the expression simplifies to
4 0 nπ $
Bn = 2 2 sin (n = 1, 2, 3, . . .).
n π 2
So
∞ 0 nπ $ . /
4 # 1 nπt
Fodd (t) = 2 sin sin .
π n2 2 T
n=1
Since f (t) and fodd (t) coincide on the interval 0 ≤ t ≤ T , this is the
required sine Fourier series Fodd (t) for the odd extension of f (t).
For n = 1, 2, 3, 4, 5, 6, 7, the values of sin(nπ/2) are 1, 0, −1, 0, 1, 0, −1, so
the first few terms in the Fourier series are
. /
4 1 1 1
Fodd (t) = 2 sin(πt) − 2 sin(3πt) + 2 sin(5πt) − 2 sin(7πt) + · · · .
π 3 5 7
Exercise 25
Consider the same function f (t) as that discussed in Example 12. Within
its domain of definition, 0 ≤ t ≤ T , represent this function by a Fourier
series that involves only constant and cosine functions.
102
Learning outcomes
Exercise 26
Consider the function
'
1 for 0 ≤ t < π2 ,
f (t) =
−1 for π2 ≤ t < π.
(a) Define the even periodic extension, simplifying your answer as much
possible. Sketch this function over −3π ≤ t ≤ 3π, and state its
fundamental period.
(b) Find the Fourier series for the even periodic extension.
(c) Define the odd periodic extension. Sketch this function over
−3π ≤ t ≤ 3π, and state its fundamental period.
Learning outcomes
After studying this unit, you should be able to:
• understand the terms frequency, period and fundamental interval, and
obtain them for a periodic function
• understand the terms even and odd as applied to functions, and test a
function to see if it is either
• find the Fourier series for a periodic function
• compare the graph of a function with the graph of a sum of terms in
the Fourier series, and comment on the closeness of the approximation
to the function
• understand how to modify a function defined on an interval to give an
even or odd periodic extension.
103
Unit 13 Fourier series
Solutions to exercises
Solution to Exercise 1
We have
1 40 1
G(t + 2π) = + 2 cos(t + 2π) + cos 3(t + 2π)
2 π 9
1 1 $
+ cos 5(t + 2π) + cos 7(t + 2π) + · · ·
25 49
1 40 1
= + 2 cos(t + 2π) + cos(3t + 6π)
2 π 9
1 1 $
+ cos(5t + 10π) + cos(7t + 14π) + · · ·
25 49
1 40 1 1 1 $
= + 2 cos t + cos 3t + cos 5t + cos 7t + · · ·
2 π 9 25 49
= G(t).
Note that G(t) would also be unchanged by adding any integer multiple of
2π to its argument. These results hold because adding 2π any number of
times to a cosine argument does not change the value of the cosine.
Solution to Exercise 2
The angular frequencies are 1, 3, 5, 7, . . .,
2π 2π 2π
and the periods are 2π, 3 , 5 , 7 , . . ..
Solution to Exercise 3
3π
The angular frequencies of the component functions are π, 2 , 2π,
4
and the corresponding periods are 2, 3, 1.
The least common multiple of these periods is 4, so the period of the
function f (t) is τ = 4.
Solution to Exercise 4
The two graphs are as follows.
104
Solutions to exercises
Solution to Exercise 5
Since g(−t) = (−t)3 = −t3 = −g(t) for al t, the function is odd.
Solution to Exercise 6
Since f (t) and g(t) are both odd functions,
f (−t) = −f (t), g(−t) = −g(t).
Hence
k(−t) = f (−t) + g(−t) = −f (t) + (−g(t)) = −(f (t) + g(t)) = −k(t),
so k(t) is an odd function.
Solution to Exercise 7
(a) The graph of the function C0 (t) is shown below. As the graph is
symmetrical about the vertical axis, the function must be even.
105
Unit 13 Fourier series
Solution to Exercise 8
We have
h(t) = f (t) g(t) = (t3 + 2t5 )(3t2 − t4 )
= 3t5 − t7 + 6t7 − 2t9
= 3t5 + 5t7 − 2t9 .
All the powers of t are odd, so h(t) is an odd function.
Alternatively, since f (t) and g(t) are odd and even functions, respectively,
we know by definition that
f (−t) = −f (t), g(−t) = g(t).
Hence
h(−t) = f (−t) g(−t) = −f (t) g(t) = −h(t),
so h(t) is an odd function.
Solution to Exercise 9
From the graph, the values of the function repeat after an interval of
length 2π. Hence the period is τ = 2π. The angular frequency ω satisfies
τ = 2π/ω, so here ω = 1.
Solution to Exercise 10
For n = 1, 2, 3, . . .,
! π ) 5π
sin nt
cos nt dt = = 0,
−π n −π
Solution to Exercise 11
Applying equation (13) gives
! π
1
A0 = h(t) dt.
2π −π
Using the hint, this simplifies to
! π/2
1
A0 = 1 dt,
2π −π/2
where we have used the fact that h(t) = 1 for −π/2 ≤ t ≤ π/2. Performing
the integration gives
1 * (π/2
A0 = t = 1.
2π −π/2 2
106
Solutions to exercises
Solution to Exercise 12
(a) As A0 is a constant, it can be taken outside the integral to obtain
! π ! π
* (π
A0 cos t dt = A0 cos t dt = A0 sin t −π = 0.
−π −π
Solution to Exercise 13
Substitute the definition of h(t) from Exercise 11 into equation (16) to
obtain
!
1 π
A1 = h(t) cos t dt.
π −π
Now use the definition of h(t) to get (as h(t) = 0 outside the interval
[− π2 , π2 ])
!
1 π/2
A1 = 1 × cos t dt.
π −π/2
Performing the integration gives
1* (π/2 1& - 2
A1 = sin t −π/2 = 1 − (−1) = .
π π π
107
Unit 13 Fourier series
Solution to Exercise 14
We could obtain a general formula for An , as in Example 6, and then
substitute n = 2 and n = 3 into that. Alternatively, put n = 2 into
equation (22) and obtain
!
1 π
A2 = h(t) cos 2t dt.
π −π
Now use the definition of h(t), which is zero outside −π/2 ≤ t ≤ π/2 and
equal to 1 on this interval, to yield
!
1 π/2
A2 = 1 × cos 2t dt.
π −π/2
Performing the integration gives
1 *1 (π/2 1 &1 -
A2 = 2 sin 2t −π/2 = 2 sin π − 12 sin(−π) = 0.
π π
Substituting n = 3 into equation (22) gives
!
1 π
A3 = h(t) cos 3t dt.
π −π
Now use the definition of h(t) again to yield
!
1 π/2
A3 = 1 × cos 3t dt.
π −π/2
Performing the integration gives
1 *1 (π/2 1 & & --
A3 = 3 sin 3t −π/2 = sin 3π
2 − sin − 3π
2
π 3π
1& - 2
= (−1) − 1 = − .
3π 3π
Solution to Exercise 15
The constant term does not fit in the general pattern, so we can leave this
term outside the summation. All of the other terms have a factor 2 in the
numerator and a factor π in the denominator, so we can factorise the
expression as
. /
1 2 1 1 1
H(t) = + cos t − cos 3t + cos 5t − cos 7t + · · · .
2 π 3 5 7
The sum in the brackets is a sum over all odd numbers. As n = 1, 2, 3, . . .,
the expression 2n − 1 evaluates to 1, 3, 5, . . ., so this is the expression to be
used in the angular frequency and the denominator. The terms also
change sign, so we need a factor (−1)n in the coefficients. The first sign is
positive, and (−1)n is negative for n = 1, so we need an extra minus sign
(either by writing this as (−1)n+1 or by taking a minus sign out of the
bracket as below).
Putting all this together gives the closed form as
∞
1 2 # (−1)n
H(t) = − cos(2n − 1)t.
2 π n=1 2n − 1
108
Solutions to exercises
Solution to Exercise 16
We have
! π ! π
1 1 1 * 1 3 (π
A0 = f (t) dt = t2 dt = t = 1 π2 ,
2π −π 2π −π 2π 3 −π 3
and for n = 1, 2, 3, . . ., we use the hint to obtain
!
1 π 2
An = t cos nt dt
π −π
) 5
1 1 & 2 2 - π
= (n t − 2) sin nt + 2nt cos nt
π n3 −π
1 1& n n
-
= × 3 2nπ(−1) − (−2nπ(−1) ) ,
π n
where we have used sin nπ = 0 and cos nπ = (−1)n as n is an integer. This
simplifies to
4(−1)n
An = .
n2
The Fourier series is therefore
#∞
π2 (−1)n
F (t) = +4 2
cos nt
3 n=1
n
π2 4 1
= − 4 cos t + cos 2t − cos 3t + cos 4t + · · · .
3 9 4
(The graph of F (t) matches the expected graph, which consists of a
parabola on the fundamental interval repeated indefinitely to both the left
and the right – see below.)
Solution to Exercise 17
The function w(t) is even, and w(t) = t on the interval [0, π]. Therefore
! π !
1 1 π 1 * 1 2 (π π
A0 = w(t) dt = t dt = t = .
2π −π π 0 π 2 0 2
For the other coefficients we must evaluate
!
1 π
An = w(t) cos nt dt.
π −π
As w(t) is even and equal to t for t positive, this simplifies to
!
2 π
An = t cos nt dt.
π 0
109
Unit 13 Fourier series
Solution to Exercise 19
Using equation (28) gives
!
1 τ/2
A0 = w(t) dt.
τ −τ/2
As w(t) is even, this integral can be simplified and evaluated as
!
2 τ/2 2 * 1 2 (τ/2 1
A0 = t dt = t = 4 τ.
τ 0 τ 2 0
110
Solutions to exercises
111
Unit 13 Fourier series
Solution to Exercise 21
(a) Using the definition of g(x), we have
f (−x) + f (−(−x)) f (−x) + f (x)
g(−x) = = = g(x).
2 2
Hence g(x) is an even function.
(b) Using the definition of h(x), we have
f (−x) − f (−(−x)) f (−x) − f (x)
h(−x) = = = −h(x).
2 2
Hence h(x) is an odd function.
(c) Using the definitions of g(x) and h(x), we have
f (x) + f (−x) f (x) − f (−x)
g(x) + h(x) = + = f (x).
2 2
Solution to Exercise 22
Using Procedure 1, we obtain
! 0 ! 1
1 1
A0 = 2 1 dt + 2 t dt
−1 0
* (0 * (1
= t −1 + 21 12 t2 0 = 12 + 41 = 34 ,
1
2
! 0 ! 1
An = cos(nπt) dt + t cos(nπt) dt
−1 0
) 5
1 * (0 1 & - 1
Using equation (18) with = sin(nπt) −1 + 2 2 cos(nπt) + nπt sin(nπt)
a = nπ. nπ n π 0
1
= 2 2 ((−1)n − 1),
n π
112
Solutions to exercises
! 0 ! 1
Bn = sin(nπt) dt + t sin(nπt) dt
−1 0
) 5
1 * (0 1 & - 1
=− cos(nπt) −1 + 2 2 sin(nπt) − nπt cos(nπt) Using equation (17) with
nπ n π 0 a = nπ.
1 1
= − (1 − cos nπ) − 2 2 (nπ cos nπ)
nπ n π
1 cos nπ cos nπ
=− + −
nπ nπ nπ
1
=− .
nπ
Solution to Exercise 23
* (
(a) It is clear from the graph of b(t) (Figure 2) that − π2 , π2 is a
fundamental interval, so b(t) has period π.
(b) The function b(t) is even, so its Fourier series involves only the
constant and cosine terms, hence
∞
#
B(t) = A0 + An cos 2nt.
n=1
113
Unit 13 Fourier series
Solution to Exercise 24
Using the definition of the even periodic extension, we have
t for 0 ≤ t ≤ 1,
3 − 1t for 1 < t ≤ 3,
qeven (t) = 2 2
−t for −1 ≤ t < 0,
3 1
2 + 2t for −3 < t < −1,
qeven (t + 6) = qeven (t).
This function has fundamental period τ = 6, and its formula cannot be
made much simpler. Its graph is sketched below.
Solution to Exercise 25
Because we are looking for a Fourier series that involves only constant and
cosine terms, we need to consider the even periodic extension of f (t),
denoted by feven (t). This is sketched in the figure below.
The function feven (t) is even and has period τ = T , so its Fourier series
takes the form
#∞ . /
2nπt
Feven (t) = A0 + An cos ,
n=1
T
where the Fourier coefficients A0 and An are given by
!
1 T /2
A0 = feven (t) dt,
T −T /2
! . /
2 T /2 2nπt
An = feven (t) cos dt (n = 1, 2, 3, . . .).
T −T /2 T
But both of the integrands are even, and on the interval 0 ≤ t ≤ T /2 we
have feven (t) = f (t) = t/T , so the integrals simplify to
!
2 T /2 t
A0 = dt,
T 0 T
! . /
4 T /2 t 2nπt
An = cos dt (n = 1, 2, 3, . . .).
T 0 T T
Evaluating the first integral gives
! T /2
2 2 * (T /2
A0 = 2 t dt = 2 12 t2 0 = 41 .
T 0 T
115
Unit 13 Fourier series
Solution to Exercise 26
(a) The even periodic extension is given by
1 for 0 ≤ t ≤ π2 ,
−1 for π < t ≤ π,
2
feven (t) =
−1 for −π < t < − π2 ,
1 for − π2 ≤ t < 0,
feven (t + 2π) = feven (t),
and is drawn below.
116
Solutions to exercises
This result is not unexpected, as the graph shows clearly that the
average value of the function is zero.
As the integrand is even, the coefficients An are given by
! π
4
An = feven (t) cos nt dt
2π 0
2! ! π 4
π/2
2
= cos nt dt − cos nt dt
π 0 π/2
2 0* (π/2 * (π $
= sin nt 0 − sin nt π/2
nπ
4 nπ
= sin .
nπ 2
Hence
∞
4#1 nπ
Feven (t) = sin cos nt.
π n=1 n 2
(c) The odd periodic extension is defined by
1 for 0 < t ≤ π2 ,
π
−1 for 2 < t < π,
fodd (t) = 1 for −π < t < − π2 ,
−1 for − π2 ≤ t < 0,
0 for t = 0 or t = π,
fodd (t + 2π) = fodd (t),
and is drawn below.
117
Unit 14
Introduction
A partial differential equation is an equation relating a dependent
variable and two or more independent variables through the partial Partial derivatives were
derivatives of the dependent variable. Differential equations have played a introduced in Unit 7.
very important role in the module so far. But until now, all the differential
equations that you have met have involved just one independent variable,
and have been equations containing one or more dependent variables and
their ordinary derivatives with respect to that independent variable. Such
equations are often called ordinary differential equations when it is Ordinary differential equations
necessary to distinguish them from partial differential equations. For many are the subject of Units 1, 6
systems that we want to be able to model, ordinary differential equations and 12.
are inadequate because the states of the system can be specified only in
terms of two – or even more – independent variables. When we are trying
to model the way in which such a system changes, we are inevitably led to
consider partial differential equations.
This unit is an introduction to partial differential equations and their
solution. The method of solution introduced here is called separation of
variables. This idea is similar to, but distinct from, the method for solving
first-order ordinary differential equations that is also called separation of
variables, described in Unit 1. Section 1 introduces partial differential
equations and concepts associated with them, then outlines the separation
of variables method that is the core of this unit.
Partial differential equations have many applications, but here we describe
just two of them. In Section 2 we look at a model of the transverse
vibrations of a taut string, such as a guitar string. First the model is The vibrations of a guitar string
derived in terms of a partial differential equation called the wave equation, are also considered in Unit 11.
then the method of separation of variables is used to find particular
solutions of the wave equation, such as the vibrations of a plucked string.
Section 3 looks at a different application, namely the modelling of the flow
of heat in a rod. This section introduces a physical law governing heat flow
called Newton’s law of cooling, then uses this law to derive a partial
differential equation modelling the flow of heat that is called the heat
equation. The section concludes by using the separation of variables
method to find particular solutions of the heat equation. A physical
phenomenon known as diffusion also satisfies the same partial differential
equation, so the heat equation is sometimes known as the diffusion
equation.
121
Unit 14 Partial differential equations
The only issue worth highlighting here is that the order of mixed
derivatives such as ∂ 2 u/∂x ∂t is 2, because we count the total number of
times that the function u is differentiated.
In this unit we consider only second-order partial differential equations,
although the methods described are applicable to partial differential
equations of any order. Furthermore, all the equations that we deal with
are linear, where again linear is defined as it is for ordinary differential
equations.
The following exercise asks you to apply these definitions to classify some
partial differential equations.
Exercise 1
We use the convention that a For each of the following differential equations, state whether it is linear or
subscript is used to denote a non-linear, and write down its order.
partial derivative with respect to
that variable. So the equation in ∂u ∂u ∂2u ∂u
(a) +u =0 (b) 2
=x +u
part (d) could be written as ∂t ∂x ∂t ∂x
∂ 3 u ∂u
+ = 0. ∂ 2u
∂x3 ∂x (c) = x2 t (d) uxxx + ux = 0
∂t ∂x
As we will see later, the wave equation is a good model of a plucked guitar
string. We will use this context to introduce other aspects of partial
differential equations. A partial differential equation model is required
because the state of the string – by which is meant its shape at any given
time after it has been plucked – requires a function u(x, t) of two
independent variables, x and t, where x is the distance along the straight
line joining the two points at which the string is anchored (which we can
consider as an axis with origin at one end of the string), and t is the time
since the string was plucked.
122
1 Solving partial differential equations
The straight line joining the two points at which the string is anchored is
the equilibrium position of the string. The dependent variable u = u(x, t)
is the transverse displacement of the string from the point on the axis
determined by x, at time t. For fixed t = t1 and varying x, u(x, t1 ) specifies
the shape of the string at time t1 , as shown in Figure 1(a). On the other
hand, we can think about a fixed x = x1 and varying t, and then u(x1 , t)
tells us how the transverse displacement of the string from the fixed point
x = x1 on the axis varies with time (see Figure 1(b)).
The model of the motion of the string that we will develop will be a
differential equation for the variable u. Since u depends on both x and t,
an equation that models the motion of the string will involve partial
derivatives of u with respect to both x and t. One such equation is
∂2u 1 ∂ 2u
2
= 2 2, (1)
∂x c ∂t
where c is a constant whose value depends on various physical
characteristics of the string. This partial differential equation is called the
wave equation. It is a very important equation of mathematical physics, The name wave equation is used
in part because it occurs in many situations that involve vibrations of because it models wave-like
extended flexible objects like strings and springs, or indeed other wave motions such as that of a
plucked guitar string.
motions such as sound waves and light waves.
Checking whether or not a given function is a solution of a given partial
differential equation is simply a matter of substituting the function into
the equation, and seeing whether it is satisfied. The only difference from
the case of an ordinary differential equation is that you have to calculate
all the relevant partial derivatives.
Example 1
Check that
u(x, t) = sin(kx) cos(kct)
is a solution of the wave equation (1) for any constant k.
123
Unit 14 Partial differential equations
Solution
We have
∂u ∂ 2u
= k cos(kx) cos(kct), = −k 2 sin(kx) cos(kct),
∂x ∂x2
∂u ∂2u
= −kc sin(kx) sin(kct), = −k 2 c2 sin(kx) cos(kct).
∂t ∂t2
So when u(x, t) = sin(kx) cos(kct),
∂2u 1 ∂ 2u
2
= 2 2,
∂x c ∂t
and this function is indeed a solution of the wave equation.
Exercise 2
Check that u(x, t) = sin(x) e−αt is a solution of the partial differential
equation
∂u ∂2u
= α 2. (2)
∂t ∂x
This partial differential equation is known as the heat equation.
124
1 Solving partial differential equations
and then releasing it. If the initial shape of the string is given by a
function f (x), then the initial conditions may be specified in the form
u(x, 0) = f (x), 0 < x < L, We use the range 0 < x < L for
the initial condition
ut (x, 0) = 0, 0 ≤ x ≤ L.
u(x, 0) = f (x) because the
The first initial condition models the initial shape of the string, while the values of u(x, 0) when x = 0 and
second corresponds to it being at rest initially. The initial condition for a x = L are specified by boundary
conditions (3).
particular initial shape is given in the following example.
Example 2
A taut string of equilibrium length L is plucked at its midpoint, which is
given an initial displacement 21 , as shown in Figure 2. It is then released
from rest.
Write down the initial conditions for the wave equation for the transverse
vibrations of this string. Figure 2 Initial position of a
taut string plucked at its
Solution
midpoint
The displacement shown in Figure 2 has two linear sections, with slopes
±1/L. Hence the initial displacement is given by
*
x/L for 0 < x ≤ 21 L,
u(x, 0) =
(L − x)/L for 12 L < x < L.
As the string is released from rest, the transverse component of the initial
velocity is given by
ut (x, 0) = 0, 0 ≤ x ≤ L.
Exercise 3
Suppose that the initial conditions for the transverse vibrations of a taut
string are
4d
− x for 0 < x ≤ 14 L,
u(x, 0) = L
− 4d (L − x) for 1 L < x < L,
3L 4
ut (x, 0) = 0, 0 ≤ x ≤ L.
Describe how the string has been set in motion.
Exercise 4
A taut string of equilibrium length L is plucked one-third of the way along
its length, which is given an initial displacement d, as shown in Figure 3.
Figure 3 Initial position of a
What is the corresponding initial condition, for this displacement, for the taut string plucked at a point
wave equation for the transverse vibrations of this string? a third of the way along its
length
125
Unit 14 Partial differential equations
Exercise 5
A taut string is initially in its equilibrium position. At time t = 0, it is
struck in such a way as to impart, instantaneously, a transverse velocity v
(in the positive direction) to the middle third of the string. This is a
simple model of a string that is hammered, such as a piano wire.
Modify the initial conditions for the wave equation for transverse
vibrations of the string to model this situation.
Example 1 showed that any function of the form u(x, t) = sin(kx) cos(kct)
is a solution of the wave equation, for any constant k. In particular,
therefore,
- . - .
πx πct
u(x, t) = sin cos (4)
L L
is a solution of the wave equation for transverse vibrations of a taut string,
where L is the equilibrium length of the string. The following exercise asks
you to show that this solution also satisfies fixed endpoint boundary
conditions and the initial condition that the string starts from rest.
Exercise 6
Show that the solution given by equation (4) satisfies the boundary
conditions
u(0, t) = u(L, t) = 0, t ≥ 0,
and the initial condition
ut (x, 0) = 0, 0 ≤ x ≤ L.
Equation (4) is not the only solution of the wave equation that satisfies the
boundary conditions. You may like to verify that each member of the
following family of functions un (x, t) (n = 1, 2, . . .) also satisfies the wave
equation, the fixed endpoint boundary conditions and the starting from
rest initial condition:
- . - .
πnx πnct
un (x, t) = sin cos . (5)
L L
The first three members of this family of solutions are shown in Figure 4 at
time t = 0.
(Note that un (x, t) denotes a family of functions indexed by the discrete
variable n; it is not the derivative with respect to n, as the notation would
imply if n were a continuous variable.)
126
1 Solving partial differential equations
You may recognise the waves shown in Figure 4 from earlier physics or
music theory as having the shape of the fundamental (n = 1) and
harmonics (n = 2, 3, . . .) of a vibrating string. You can observe that the
larger the value of n, the higher the frequency (and the higher the pitch of
the note).
In music we rarely hear pure tones such as the ones shown in Figure 4 as
musical instruments each produce a characteristic mixture of pure tones
that gives the characteristic sound (known as the timbre) of an instrument.
In terms of a mathematical model of a taut string, the mixing of tones
corresponds to taking linear combinations of the family of solutions (5).
This is a key feature of the type of partial differential equations that we
study, namely linear differential equations that are homogeneous, in that
each additive term involves the dependent variable or its derivatives –
there are no constant terms or terms involving solely the independent
variables. The key result is as follows.
Principle of superposition
If u and v are solutions of a linear homogeneous partial differential
equation, then Au + Bv is also a solution of the same equation for
any constants A and B. Furthermore, if u and v also satisfy
homogeneous boundary conditions such as u(0, t) = 0 or ux (L, t) = 0,
then the linear combination Au + Bv will also satisfy them.
127
Unit 14 Partial differential equations
Exercise 7
Consider the heat equation
∂Θ ∂ 2Θ
=α ,
∂t ∂x2
subject to boundary conditions
Θ(0, t) = Θ(L, t) = Θ0 , t ≥ 0,
where Θ0 *= 0.
(a) Show that the function Θ(x, t) = Θ0 satisfies the differential equation
and boundary conditions.
(b) Define u(x, t) = Θ(x, t) − Θ0 . Show that u(x, t) satisfies the heat
equation with homogeneous boundary conditions.
Exercise 8
Consider the heat equation
∂Θ ∂ 2Θ
=α ,
∂t ∂x2
subject to boundary conditions
Θ(0, t) = Θ0 , Θ(L, t) = ΘL , t ≥ 0,
where Θ0 and ΘL are non-zero constants.
(a) Show that the function
L−x x
Θ(x, t) = Θ0 + ΘL
L L
satisfies the differential equation and the boundary conditions.
(b) Define the function
L−x x
u(x, t) = Θ(x, t) − Θ0 − ΘL .
L L
Show that u(x, t) satisfies the heat equation with homogeneous
boundary conditions.
128
1 Solving partial differential equations
Exercise 9
If the function u is defined as a product
u(x, t) = X(x) T (t),
find formulas for the partial derivatives ∂u/∂t, ∂ 2 u/∂t2 , ∂u/∂x and
∂ 2 u/∂x2 in terms of the functions X and T and their ordinary derivatives.
129
Unit 14 Partial differential equations
Exercise 10
Consider the differential equation
∂2u ∂u
+ 2u = .
∂x2 ∂t
Use the substitution u(x, t) = X(x) T (t) to separate the variables x and t,
and hence find the two differential equations satisfied by X(x) and T (t).
The form of the solution of the equation involving x depends on the value
of the separation constant µ and splits into three cases.
Note that the constants c and k that appear in the box above are real
numbers (in the cases where they apply) and are positive (since the square
root function gives the positive root).
Now we turn our attention to the boundary conditions for the partial
differential equation, which can be used to derive boundary conditions for
the separated ordinary differential equations. As an example to show the
general method, consider the boundary condition u(0, t) = 0 for t ≥ 0.
Substituting u(x, t) = X(x) T (t) into the boundary condition gives
X(0) T (t) = 0.
130
1 Solving partial differential equations
This equation implies that either X(0) = 0 or T (t) = 0 for all t. The latter
option gives u(x, t) = X(x) × 0 = 0, which is known as the trivial solution
(it is always a solution of a linear homogeneous differential equation). So
for non-trivial solutions of the partial differential equation we must have
X(0) = 0. Hence the boundary condition for u(x, t) imposes a boundary
condition on X(x).
Similar results hold for other boundary conditions, as the next exercise
shows.
Exercise 11
Consider the boundary condition ux (1, t) = 0 for t ≥ 0. If
u(x, t) = X(x) T (t), then what boundary condition must X(x) satisfy in
order to find non-trivial solutions of a partial differential equation?
131
Unit 14 Partial differential equations
132
1 Solving partial differential equations
133
Unit 14 Partial differential equations
The next exercise asks you to follow the first two steps of this procedure to
find a family of normal mode solutions for a different partial differential
equation.
Exercise 12
Use Procedure 1 to find an infinite family of normal mode solutions for the
partial differential equation
∂2u ∂2u
This equation is known as + 2 = 0,
Laplace’s equation. ∂x2 ∂t
subject to boundary conditions
u(0, t) = u(1, t) = 0, t ≥ 0.
134
2 The wave equation
Figure 6 Splitting the string into small segments: imagine the string
being split at the dashed lines and replaced by particles located at the
midpoints
135
Unit 14 Partial differential equations
M δx/L located at the point (x, u(x, t)), as shown in Figure 7. Also shown
in Figure 7 is the angle θ(x, t), which is the angle that the tension force
due to the string makes to the left of the particle at position x at time t.
j
i
Also shown in Figure 7 are the forces acting on the three segments. By
assumption (c), all forces are negligible compared to the tension in the
string, so we have modelled only the tension forces. As the tension forces
between neighbouring particles (e.g. T4 and T1 ) are due to a portion of
string at a fixed angle, they must be of equal magnitude and opposite
direction, as shown. The total force on the central segment at position x is
T1 + T2 .
By assumption (b), the central segment moves vertically, so its acceleration
is utt (x, t) j. Applying Newton’s second law gives
M δx
utt (x, t) j = T1 + T2 . (11)
L
We now need expressions for T1 and T2 . Resolving in the i-direction gives
0 = T1 · i + T2 · i.
From this equation we get |T1 · i| = |T2 · i|, that is, the magnitude of the
horizontal component of the tension force is constant. As the string is
horizontal in equilibrium, this is equal to the equilibrium tension, Teq .
This gives the diagram shown in Figure 8.
M δx
Teq L Teq tan(θ(x + δx, t))
Figure 8 Resolving the forces in the central line segment (vertical scale
exaggerated for clarity)
136
2 The wave equation
Wave equation
∂2u 1 ∂2u
= . (14)
∂x2 c2 ∂t2
Example 3
Use equation (14) to determine the dimensions of the constant c.
137
Unit 14 Partial differential equations
Solution
The dimensions of a derivative [∂ 2 u/∂x2 ] = L L−2 = L−1 ,
were dealt with in Exercise 21 of
[∂ 2 u/∂t2 ] = L T−2 .
Unit 8.
Therefore
[∂ 2 u/∂t2 ] L T−2
[c2 ] = = = L2 T−2 ,
[∂ 2 u/∂x2 ] L−1
so [c] = L T−1 .
Exercise 13
#
Show that Teq L/M has the dimensions of speed.
138
2 The wave equation
Example 4
In this example we consider the vibrations of a guitar string of length L
that is held fixed at its endpoints. The string is plucked, that is, released
from rest with an initial profile given by a function f (x), 0 < x < L.
We model the guitar string as a taut string, so the transverse displacement
of the string at position x and time t will satisfy the wave equation
∂2u 1 ∂ 2u
= . (16)
∂x2 c2 ∂t2
The fixed endpoints of the string give the boundary conditions
u(0, t) = 0 and u(L, t) = 0, t ≥ 0. (17)
The initial profile of the string gives the initial condition
u(x, 0) = f (x), 0 < x < L, (18)
and the fact that it is released from rest gives the initial condition
ut (x, 0) = 0, 0 ≤ x ≤ L. (19)
Find u(x, t) that satisfies this model.
Solution
Applying Procedure 1, we look for solutions of the form " Separate variables !
u(x, t) = X(x) T (t). Here T is a function of time. It
should not be confused with the
Differentiating this expression gives magnitude of a force T or the
∂u ∂ 2u dimension T.
= X(x) T ! (t), = X(x) T !! (t),
∂t ∂t2
∂u ∂ 2u
= X ! (x) T (t), = X !! (x) T (t).
∂x ∂x2
Substituting into equation (16) gives
1
X !! (x) T (t) =X(x) T !! (t),
c2
which on rearranging gives
X !! (x) T !! (t)
= 2 .
X(x) c T (t)
Since the left-hand side is a function of x only, and the right-hand side is a
function of t only, both sides must be constant and equal to a separation
constant µ. This gives the two equations
X !! (x) T !! (t)
= µ and = µ.
X(x) c2 T (t)
Rearranging gives two ordinary differential equations:
X !! (x) = µ X(x), T !! (t) = µc2 T (t).
The given boundary conditions (17) become
X(0) T (t) = 0 and X(L) T (t) = 0, t ≥ 0.
139
Unit 14 Partial differential equations
140
2 The wave equation
141
Unit 14 Partial differential equations
142
2 The wave equation
Figure 10 (a) Snapshots of the profile of the string initially (red), then
after successive tenths of a cycle, green, brown, purple, orange, blue (and
then back again). (b) The solution u(x, t) of the damped wave equation
plotted as a surface.
143
Unit 14 Partial differential equations
This model predicts that the fundamental angular frequency of the guitar
string will be the smallest coefficient of t in the cosine terms of the series,
that is, πc/L. The fundamental frequency f is obtained by dividing the
fundamental angular frequency by 2π:
ω πc/L c
f= = = .
2π 2π 2L
Now c is related to the properties of the string by the relationship (13),
namely c2 = Teq L/M . This can be substituted into the above equation to
yield
1 1 1
1 Teq L 1 Teq 1 68
f= = = = 323.4.
2L M 2 ML 2 0.25 × 10−3 × 0.65
So this continuous model of the guitar string gives an accurate prediction
of the fundamental frequency.
The following exercise looks at a different initial condition.
Exercise 14
Write down the solution for the plucked string model when the initial
profile f (x) is given by
- .
1 3πx
f (x) = sin .
2 L
Exercise 15
Consider the damped wave equation involving the constants ε and c,
- .
∂2u 1 ∂2u ∂u
= 2 + 2ε ,
∂x2 c ∂t2 ∂t
together with the boundary conditions
u(0, t) = u(L, t) = 0, t ≥ 0.
Apply the first step of Procedure 1 to separate the variables and obtain
two ordinary differential equations together with corresponding boundary
conditions.
144
2 The wave equation
The results of this exercise are used in the following example, which uses
arguments about the relative sizes of parameters to find an approximate
solution to the problem. This often happens with complicated
mathematical models – either the exact equations cannot be solved or the
solution is too complicated to give insight into the problem – and
consequently an approximate solution to the problem is sought. This
example is consequently harder than the other examples in this unit.
Example 5
Consider the model developed in Subsection 2.1 for damped vibrations of a
string,
- .
∂2u 1 ∂2u ∂u
= 2 + 2ε , (22)
∂x2 c ∂t2 ∂t
where ε and c are constants. In this example we consider weak damping,
which is the case when ε is small.
As in Example 4, the string is subject to fixed endpoint boundary
conditions
u(0, t) = u(L, t) = 0, t ≥ 0, (23)
and initial conditions that the string starts from rest in a given shape:
*
x/L for 0 < x ≤ 21 L,
u(x, 0) = (24)
(L − x)/L for 12 L < x < L,
ut (x, 0) = 0, 0 ≤ x ≤ L. (25)
Solution
Using the results of Exercise 15, the differential equations corresponding to " Separate variables !
the given partial differential equation are
X !! (x) − µ X(x) = 0 (26)
and
T !! (t) + 2ε T ! (t) − c2 µ T (t) = 0, (27)
where µ is the separation constant. In addition, we have
X(0) = 0 and X(L) = 0, (28)
which are boundary conditions for the ordinary differential equation (26).
The differential equation for X, and its boundary conditions, are the same " Solve ODEs !
as in Example 4. You have seen that a non-trivial solution occurs only if
the separation constant µ is negative, and is given by
X(x) = A cos kx + B sin kx,
√
where A and B are constants, and k = −µ.
145
Unit 14 Partial differential equations
146
2 The wave equation
147
Unit 14 Partial differential equations
This is the same as the solution to the undamped problem except for the
factor e−εt , which describes exponential decay. Hence the shape of the
string is roughly the same as before, apart from the amplitude becoming
progressively smaller by the factor e−εt . The shape determines the sound,
so the sound stays the same, only now it gets progressively quieter, since
the volume is determined by the amplitude. The solution is shown in
Figure 11.
The figure shows that the exponential factor has the effect of reducing the
amplitude of the vibrations of the string, which is to be expected for a
solution of the damped wave equation.
Figure 11 The solution
u(x, t) of the damped wave
equation shown over two
complete cycles
148
3 The heat equation
This law applies to objects that are not changing state in the process (such
as a solid melting into a liquid). Although Newton called this a law of
nature, it is really just a first model of how heat flows. It should be
considered in a similar way to Hooke’s law for springs, where we used the
law to derive useful models of oscillating systems.
We will use Θ to denote the temperature, so this law can be formulated as
dΘ
= −k(Θ − Θ0 ),
dt
where Θ0 is the temperature of the surrounding environment, and k is a
positive constant of proportionality. The negative sign in this equation is
because hotter objects (i.e. with Θ > Θ0 ) cool down (so dΘ/dt < 0). The
constant of proportionality k depends on many properties of the object
(such as size and mass) and the contact between the object and its
surroundings (such as whether there is direct contact or an air gap), but
will be constant for a given object in a given situation.
The study of the flow of heat is called thermodynamics, and this is a
substantial topic in modern physics. Here we are not concerned with the
details of thermodynamics and the different methods of heat transfer. We
will use Newton’s law of cooling as a mathematical model of how everyday
physical objects behave in an analogous way to how we use Hooke’s law as
a mathematical model to describe the behaviour of physical springs.
Armed with this mathematical model, we can derive a partial differential
equation that models the change in temperature distribution in a uniform
rod as it conducts heat. We will assume that no heat is lost from the sides
of the rod and that heat is transferred only along the length of the rod (so
that we have a one-dimensional problem). We are modelling a straight rod,
and we choose an x-axis to be aligned with the rod with the origin at the
left-hand end. Let Θ(x, t) be the temperature at position x along the rod
at time t. Consider a small segment of rod of length δx with midpoint a
distance x along the rod, as shown in Figure 12.
We will now apply our model of cooling with the small central segment of
the rod considered as the object. The temperature on the left of the
central segment is Θ(x − δx, t) and this takes the role of the temperature
of the environment for the interface between the central and left-hand Figure 12 A conducting rod
segments,
& so the temperature
, change due to this left-hand interface is
−k Θ(x, t) − Θ(x − δx, t) . Similarly, on the right-hand end of the central
segment, the ‘environment’ temperature is Θ(x + δx, t), and the
temperature
& change due to
, heat exchange at this interface is
−k Θ(x, t) − Θ(x + δx, t) (note that the constant of proportionality k is
the same for both ends as the rod is uniform thus the ends are identical).
The change in temperature of the central segment Θt (x, t) is given by the
sum of these two contributions:
& , & ,
Θt (x, t) = −k Θ(x, t) − Θ(x − δx, t) − k Θ(x, t) − Θ(x + δx, t)
& ,
= k Θ(x − δx, t) − 2Θ(x, t) + Θ(x + δx, t) . (34)
149
Unit 14 Partial differential equations
Heat equation
∂Θ ∂ 2Θ
=α . (35)
∂t ∂x2
If the ends of the rod are kept at a steady temperature Θ0 , the boundary
conditions are
Θ(0, t) = Θ(L, t) = Θ0 , t ≥ 0. (36)
Further, if the initial distribution of temperature along the rod is given by
Note that we need only one a function f (x), then we have the initial condition
initial condition because the
equation is first-order in time. Θ(x, 0) = f (x), 0 < x < L. (37)
These conditions are sufficient for us to be able to obtain a unique solution
of the heat equation for the variation of temperature in an insulated rod,
as you will see in Subsection 3.2.
Exercise 16
Show that the function
/ πx $ - .
απ2 t
Θ(x, t) = sin exp − 2
L L
satisfies the heat equation (35) and the boundary conditions (36) if Θ0 = 0.
Exercise 17
Suppose that initially, the temperature of a rod rises linearly with x
towards a peak in the centre that is half a degree above the end
temperature Θ0 , as shown in Figure 13.
Figure 13 An initial
Write down a formula describing the initial temperature function f (x).
temperature distribution
150
3 The heat equation
Exercise 18
Write down the initial condition describing the temperature distribution if
the central third of a rod is initially heated to a temperature Θ1 while the
remainder of the rod stays at the background temperature Θ0 .
Example 6
Apply Procedure 1 to the heat equation
∂Θ ∂ 2Θ
=α ,
∂t ∂x2
subject to the boundary conditions These boundary conditions
model the situation where the
Θx (0, t) = Θx (L, t) = 0, t ≥ 0, ends of a hot rod are insulated.
and the initial condition
- .
1 2πx
Θ(x, 0) = 2 cos , 0 < x < L.
L
151
Unit 14 Partial differential equations
Solution
" Separate variables ! We write Θ(x, t) = X(x) T (t), so
∂Θ ∂ 2Θ ∂Θ
= X ! T, 2
= X !! T and = XT ! .
∂x ∂x ∂t
Substituting into the partial differential equation gives XT ! = αX !! T .
Separate the variables to yield
X !! T!
= .
X αT
Arguing as before, both sides of this equation must be equal to a
separation constant µ, so we have the two differential equations
X !! − µX = 0 and T ! − αµT = 0.
We have ∂Θ/∂x = X ! T , so the boundary conditions become
X ! (0) T (t) = X ! (L) T (t) = 0, t ≥ 0,
hence
X ! (0) = X ! (L) = 0.
" Solve ODEs ! Consider the three cases µ = k 2 > 0, µ = 0 and µ = −k 2 < 0.
√
• µ > 0. Let c = µ. Then equation (9) gives the solution of the
differential equation for X(x) as
X(x) = Aecx + Be−cx ,
where A and B are constants.
Now we apply the boundary conditions, and to do this we first
differentiate this solution:
X ! (x) = Acecx − Bce−cx .
The boundary condition X ! (0) = 0 gives Ac − Bc = 0, so A = B as
c > 0. The boundary condition X ! (L) = 0 gives
AcecL − Bce−cL = 0.
As A = B and c > 0, this simplifies to
A(ecL − e−cL ) = 0.
Proceeding as before we can multiply by ecL to obtain
A(e2cL − 1) = 0.
Now we can see that the term in brackets is not zero as 2cL > 0 so
exp(2cL) > exp(0) = 1. So we have A = B = 0, and the only solution
is the trivial solution.
• µ = 0. In this case the differential equation for X(x) becomes X !! = 0,
with solution X(x) = Ax + B, where A and B are constants. So
X ! (x) = A, and both boundary conditions give the equation A = 0.
So the solution X(x) = B is a non-trivial solution of the equation.
152
3 The heat equation
√
• µ < 0. Let k = −µ. Then by equation (9) the general solution of the
equation for X is
X(x) = A cos kx + B sin kx,
where A and B are constants, so
X ! (x) = −Ak sin kx + Bk cos kx.
Using the boundary conditions, we find that B = 0, and A = 0 or
k = nπ/L for any non-zero integer n. This leads to the solution
/ nπx $
X(x) = A cos , n = 1, 2, 3, . . . .
L
So we have two cases that give non-trivial solutions, namely the constant
solutions that correspond to µ = 0 and the sinusoidal solutions that
correspond to µ < 0. Here these two cases can be conveniently combined
by adding n = 0 to the second set of solutions (using the fact that
cos(0 × x) = 1 for all x). So we have the solutions
/ nπx $
X(x) = A cos , n = 0, 1, 2, 3, . . . .
L
With µ = n2 π2 /L2 , the differential equation for T (t) becomes
n2 π2
T ! (t) + α T (t) = 0.
L2
Equation (8) gives the solution of this equation as
- .
αn2 π2 t
T (t) = C exp − , n = 0, 1, 2, 3, . . . ,
L2
where C is a constant.
This leads to the family of solutions
- . / nπx $
αn2 π2 t
Θn (x, t) = an exp − cos , n = 0, 1, 2, 3, . . . ,
L2 L
where the an = AC are constants.
To find the solution that satisfies the initial condition, write " Initial conditions !
"∞ - . / nπx $
αn2 π2 t
Θ(x, t) = an exp − cos , (39)
n=0
L2 L
and set t = 0 to give
∞
" / nπx $
Θ(x, 0) = an cos .
n=0
L
Now we use the given initial condition
- .
2πx
Θ(x, 0) = 21 cos .
L
By inspection we have a2 = 12 , and an = 0 for n *= 2, so the solution is
- . - .
1 4απ2 t 2πx
Θ(x, t) = 2 exp − 2 cos .
L L
153
Unit 14 Partial differential equations
Exercise 19
Solve the heat equation
∂Θ ∂ 2Θ
=α , (40)
∂t ∂x2
These boundary conditions subject to the boundary conditions
model the situation where the
ends of the rod are held at a Θ(0, t) = Θ(L, t) = 0, t ≥ 0, (41)
fixed temperature.
and the initial condition
/ πx $
Θ(x, 0) = sin . (42)
L
Exercise 20
Find the solution of the uninsulated rod equation
∂Θ ∂ 2Θ
This is equation (38) derived in =α − γΘ, (43)
Subsection 3.1, with Θ0 = 0. ∂t ∂x2
subject to the boundary conditions
Θ(0, t) = Θ(L, t) = 0, t ≥ 0, (44)
and the initial condition
/ πx $
Θ(x, 0) = sin . (45)
L
Compare the solution with the solution to the insulated rod problem in
Example 6.
154
Learning outcomes
Learning outcomes
After studying this unit, you should be able to:
• use the terms linear, homogeneous, order, initial condition and
boundary condition as applied to partial differential equations
• show that a given function satisfies a given partial differential equation
and/or boundary conditions and/or initial conditions
• use the method of separation of variables to find solutions of linear
homogeneous second-order partial differential equations
• understand how the wave equation and heat equation can be used to
model certain physical systems
• interpret solutions of partial differential equations in terms of a model.
155
Unit 14 Partial differential equations
Solutions to exercises
Solution to Exercise 1
(a) This equation is non-linear (because of the term that contains a
product of u and ux ). It is a first-order equation.
(b) This equation is a linear second-order equation.
(c) This equation is a linear second-order equation.
(d) This equation is a linear third-order equation.
Solution to Exercise 2
Start by differentiating u(x, t):
∂u ∂2u ∂u
= cos(x) e−αt , = − sin(x) e−αt , = −α sin(x) e−αt .
∂x ∂x2 ∂t
So
∂ 2u ∂u
α 2
= −α sin(x) e−αt = ,
∂x ∂t
and the partial differential equation is satisfied.
Solution to Exercise 3
The initial velocity is zero, so the string is released from rest. When
x = 14 L, u(x, 0) = −d, so the point one-quarter of the way along the string
has been displaced downwards by a distance d.
Solution to Exercise 4
The initial displacement has two linear sections, with slopes
d 3d d 3d
1 = and −2 =− ,
3L
L 3L
2L
respectively. Hence the required initial condition is
3d
x for 0 < x ≤ 13 L,
u(x, 0) = L
3d (L − x) for 1 L < x < L.
2L 3
Solution to Exercise 5
Initially, there is no displacement, so
u(x, 0) = 0, 0 < x < L.
Since only the middle third is set in motion, the initial velocity is given by
'
v for 31 L ≤ x ≤ 32 L,
ut (x, 0) =
0 otherwise.
156
Solutions to exercises
Solution to Exercise 6
The boundary conditions are satisfied since
- .
πct
u(0, t) = sin 0 cos = 0,
L
- .
πct
u(L, t) = sin π cos = 0.
L
The initial condition is satisfied since
πc / πx $
ut (x, 0) = − sin sin 0 = 0.
L L
Solution to Exercise 7
(a) Since the function Θ(x, t) = Θ0 is constant, all its derivatives are zero,
and the differential equation reduces to 0 = 0 thus is satisfied.
As Θ(x, t) is always equal to Θ0 , in particular it is equal to Θ0 at the
boundaries, that is, the boundary conditions are satisfied.
(b) As Θ0 is constant, its derivatives are zero, so ut = Θt and uxx = Θxx .
So if Θ satisfies the heat equation, then so does u.
On the boundary we have Θ(0, t) = Θ0 , so u(0, t) = Θ0 − Θ0 = 0.
Similarly, Θ(L, t) = Θ0 , so u(L, t) = Θ0 − Θ0 = 0.
So u(x, t) satisfies the heat equation with homogeneous boundary
conditions.
Solution to Exercise 8
(a) For
L−x x
Θ(x, t) = Θ0 + ΘL ,
L L
we have
∂ 2Θ ∂Θ
2
= 0 and = 0,
∂x ∂t
so the differential equation reduces to 0 = 0 thus is satisfied.
When x = 0 we have
L−0
Θ(0, t) = Θ0 + 0 = Θ0 ,
L
and when x = L we have
L−L L
Θ(L, t) = Θ0 + ΘL = ΘL ,
L L
so the boundary conditions are satisfied.
157
Unit 14 Partial differential equations
Solution to Exercise 9
∂u ∂ 2u
= X(x) T ! (t), = X(x) T !! (t),
∂t ∂t2
∂u ∂ 2u
= X ! (x) T (t), = X !! (x) T (t).
∂x ∂x2
Solution to Exercise 10
Proceeding as in the main text, we use the derivatives found in Exercise 9,
namely
∂ 2u ∂u
2
= X !! (x) T (t) and = X(x) T ! (t).
∂x ∂t
Substituting these into the partial differential equation gives
X !! (x) T (t) + 2 X(x) T (t) = X(x) T ! (t).
Dividing by X(x) T (t) gives
X !! (x) T ! (t)
We chose to leave the 2 on the +2= .
left-hand side. It would be X(x) T (t)
equally correct to write the This has separated the variables, so both sides must be equal to a
separated equations as constant, say µ. So we have the two equations
X !! (x) T ! (t)
= − 2. X !! (x) T ! (t)
X(x) T (t) + 2 = µ and = µ.
X(x) T (t)
Multiplying out the fractions and rearranging gives the required
differential equations:
X !! (x) + (2 − µ) X(x) = 0 and T ! (t) − µ T (t) = 0.
158
Solutions to exercises
Solution to Exercise 11
As the boundary condition is defined by the partial derivative with respect
to x, we first find
∂u
= X ! (x) T (t).
∂x
So the boundary condition becomes X ! (1) T (t) = 0 for t ≥ 0. The solution
T (t) = 0 for t ≥ 0 leads to the trivial solution. So for a non-trivial solution
we must have X ! (1) = 0.
Solution to Exercise 12
Setting u(x, t) = X(x) T (t), the required partial derivatives are " Separate variables !
∂ 2u !! ∂ 2u
= X T and = XT !! .
∂x2 ∂t2
Substituting into the partial differential equation and dividing by XT gives
X !! T !!
+ = 0,
X T
from which it follows that
X !! T !!
=− .
X T
Both sides of this equation must be a constant, say µ, giving
X !! T !!
= µ and − =µ
X T
or equivalently,
X !! − µX = 0 and T !! + µT = 0.
The boundary conditions become X(0) = X(1) = 0.
The boundary conditions are the same as those used in the main text. So " Solve ODEs !
arguing as in the text, only negative µ gives a non-trivial solution for X.
In this case, the equation for X has general solution (see equation (9))
X(x) = A cos kx + B sin kx,
√
where k = −µ, and A and B are constants. The boundary condition
X(0) = 0 implies A = 0. The boundary condition X(1) = 0 implies
B sin k = 0, so for non-trivial solutions we must have sin k = 0, that is,
k = nπ for some positive integer n. For this value of k we have
X(x) = B sin(nπx), n = 1, 2, 3, . . . .
Since µ = −n2 π2 , the equation for T can be written as
T !! (t) − n2 π2 T (t) = 0.
Using equation (9), the solutions in this case are (as n2 π2 is positive)
T (t) = Cenπt + De−nπt , n = 1, 2, 3, . . . ,
where C and D are constants.
159
Unit 14 Partial differential equations
Solution to Exercise 13
Teq is a component of a force, so [Teq ] = M L T−2 . Also, [M ] = M and
[L] = L. Hence
[Teq L/M ] = M L T−2 × L/M = L2 T−2 ,
#
so [ Teq L/M ] = L T−1 , as required.
Solution to Exercise 14
To find the solution we need to find the coefficient an appearing in the
equation
"∞ - . - .
nπx 1 3πx
an sin = sin .
n=1
L 2 L
This can be done by inspection as the term on the right-hand side appears
as one of the terms on the left-hand side. So we get a3 = 21 , and an = 0 for
n *= 3.
So the solution for the model is given by equation (21) with the above
coefficients, that is,
- . - .
1 3πx 3cπt
u(x, t) = sin cos .
2 L L
Solution to Exercise 15
Write the unknown function as
u(x, t) = X(x) T (t).
Differentiate to get the partial derivatives:
∂u ∂ 2u ∂ 2u
= X(x) T ! (t), = X(x) T !!
(t), = X !! (x) T (t).
∂t ∂t2 ∂x2
Substituting these into the damped wave equation gives
1
X !! (x) T (t) = 2 (X(x) T !! (t) + 2ε X(x) T ! (t)) .
c
As before, this equation can hold for all x and all t only if both sides are
equal to a separation constant µ, that is,
- .
X !! (x) 1 T !! (t) T ! (t)
= 2 + 2ε = µ,
X(x) c T (t) T (t)
giving a pair of equations.
160
Solutions to exercises
Solution to Exercise 16
/ πx $ - .
∂Θ π απ2 t
= cos exp − 2 ,
∂x L L L
so
/ πx $ - .
∂ 2Θ π2 απ2 t π2
= − sin exp − = − Θ(x, t).
∂x2 L2 L L2 L2
Also,
/ πx $ - .
∂Θ απ2 απ2 t απ2
= − 2 sin exp − 2 = − 2 Θ(x, t).
∂t L L L L
Thus
∂Θ π2 ∂ 2Θ
= −α 2 Θ(x, t) = α .
∂t L ∂x2
Hence equation (35) is satisfied.
The boundary conditions are
- .
απ2 t
Θ(0, t) = exp − 2 sin 0 = 0,
L
- .
απ2 t
Θ(L, t) = exp − 2 sin π = 0,
L
so these are satisfied for a temperature at the rod ends of Θ0 = 0.
161
Unit 14 Partial differential equations
Solution to Exercise 17
The graph in Figure 13 looks just like the picture of the plucked string in
Figure 2, so the required formula is
*
Θ0 + x/L for 0 < x ≤ 21 L,
f (x) =
Θ0 + (L − x)/L for 21 L < x < L.
As required, this function takes the value Θ0 when x is 0 or L, and takes
the value Θ0 + 12 when x is 21 L.
Solution to Exercise 18
Θ0
for 0 < x < 13 L,
Θ(x, 0) = Θ1 for 31 L ≤ x ≤ 32 L,
Θ0 for 23 L < x < L.
Solution to Exercise 19
" Separate variables ! We begin as in Example 6, and obtain the equations
X !! − µX = 0 and T ! − αµT = 0.
To find boundary conditions for X, we put x = 0 and x = L in
Θ(x, t) = X(x) T (t) and substitute into the boundary conditions (41),
which gives
X(0) T (t) = X(L) T (t) = 0, t ≥ 0,
hence
X(0) = X(L) = 0.
" Solve ODEs ! Next we solve the differential equations for X and T , and combine the
families of solutions.
The differential equation for X, and its boundary conditions, are the same
as in Example 4. You have seen that a non-trivial solution occurs only if
the separation constant µ is negative, and is given by
/ nπx $
X(x) = B sin , n = 1, 2, 3, . . . ,
L
where B is a constant. In this case the separation constant µ is −n2 π2 /L2 .
Our next task is to solve the equation for T when µ = −n2 π2 /L2 , namely
αn2 π2
T ! (t) + T (t) = 0.
L2
The general solution is given by equation (8) as
- .
αn2 π2 t
T (t) = C exp − ,
L2
where C is a constant.
162
Solutions to exercises
Solution to Exercise 20
Set Θ(x, t) = X(x) T (t). Then " Separate variables !
∂ 2Θ ∂Θ
2
= X !! T and = XT ! .
∂x ∂t
Equation (43) becomes
XT ! = αX !! T − γXT,
and dividing by XT gives
T! X !!
=α − γ.
T X
This can be rearranged as
- .
1 T! X !!
+γ = .
α T X
Again, a function of x is equal to a function of t, so both must be constant.
Choosing the constant to be µ = −k 2 , as before, the equations become
X !! + k 2 X = 0 and T ! + (αk 2 + γ)T = 0.
The boundary conditions reduce to
X(0) = 0 and X(L) = 0.
163
Unit 14 Partial differential equations
" Solve ODEs ! Solving the differential equation for X(x) subject to fixed endpoint
boundary conditions leads again to k = nπ/L, for any positive integer n,
and then to the family of solutions for X given by
/ nπx $
X(x) = B sin , n = 1, 2, 3, . . . ,
L
where B is a constant.
The solution for T is given by equation (8) with k = nπ/L:
- - 2 2 . .
αn π
T (t) = A exp − + γ t , n = 1, 2, 3, . . . ,
L2
where A is a constant.
So the family of solutions is
- - 2 2 . . / nπx $
αn π
Θn (x, t) = an exp − + γ t sin , n = 1, 2, 3, . . . ,
L2 L
where the an = BA are constants.
" Initial conditions ! Now use the principle of superposition to write down a more general
solution:
"∞ - - 2 2 . . / nπx $
αn π
Θ(x, t) = an exp − + γ t sin .
n=1
L2 L
Setting t = 0 gives
/ πx $ "∞ / nπx $
sin = Θ(x, 0) = an sin .
L n=1
L
By inspection, a1 = 1, and an = 0 for n = 2, 3, . . ., as the term on the
left-hand side is one member of the sum on the right-hand side. Therefore
the solution is
- - 2 . . / πx $
απ
Θ(x, t) = exp − + γ t sin
L2 L
- 2
. / $
απ t πx
= e−γt exp − 2 sin .
L L
This is the same as the solution for the insulated rod except for the e−γt
factor, so the uninsulated rod cools more quickly than the insulated rod,
by a factor of e−γt .
164
Index
Index
angular frequency 66 limit cycle 43
linear partial differential equation 122
boundary conditions 124 linearising a system near an equilibrium point 20
Lotka, Alfred 10
centre 25
Lotka–Volterra equations 10
classification of an equilibrium point 28, 32
closed form 81 Newton’s law of cooling 148
damped wave equation 138, 144 normal mode solutions 133
diffusion equation 121 nullcline 37
direction field 5
odd function 68
eigenline 22 odd periodic extension 98
equilibrium point 13 odd periodic function 71, 88
centre 25 orbit 7
classification of 29 order of a partial differential equation 122
improper sink 28 ordinary differential equation 121
improper source 28
linearising a system near 20 partial differential equation 121
saddle 8, 24 path 5
sink 23 period (fundamental) 66
source 22 periodic extension 97
spiral sink 26 periodic function 66, 94
spiral source 27 phase path 7
stability of 15, 16 phase plane 7
star sink 28 phase portrait 7, 39
star source 28 pointwise convergence theorem 83
even function 68 predator–prey model 8
even periodic extension 98 principle of superposition 127
even periodic function 71, 85 procedure
equilibrium point classification for a linear system
Fourier, Joseph 64 29
Fourier coefficients 73 equilibrium point classification for a non-linear
Fourier series system 32
description 64 finding equilibrium points 14
for a function on a finite interval 97 Fourier series for periodic functions 94
procedure 94 linearising a system of differential equations near an
fundamental interval 67 equilibrium point 20
fundamental period 66 separation of variables 134
sketching phase portraits 39
growth rate 5
saddle point 8, 24
heat equation 124, 150 sawtooth function 63
homogeneous partial differential equation 127 separation constant 129
separation of variables 129, 134
improper sink 28
sink 8, 23
improper source 28
source 7, 22
initial conditions 124
spiral sink 26
insulated rod 148
spiral source 27
integrals of even and odd periodic functions 72
square-wave function 63
Jacobian matrix 19 stable equilibrium point 15, 16, 23, 25, 26, 28
star sink 28
Laplace’s equation 134 star source 28
165
Index
166